Forty-four jurisdictions can reopen a completed deal. Four give the buyer a date that runs from the transaction itself and expires. Twenty-seven give no date at all.
Conflicts & Capital is a column on the private international law of cross-border deals, written by Eric Martin for CICL Currents at Emory Law. Start with the Brief, or put two countries into the Path Finder and see whether a judgment between them travels.
The floor for an own-initiative review after closing, where the investment was not subject to a prior authorization requirement. Member States may elect up to five years. A deal that was notifiable and was never filed carries a 24-month floor and no ceiling at all, under art. 4(5).
Reg (EU) 2026/1386, art. 4(4)
27
Member States that must operate a screening mechanism once the Regulation applies on January 17, 2028.
Reg (EU) 2026/1386 · minimum scope, not full harmonization
30
Of the 116 entries in Release 01, the number party to the New York Convention and to nothing else in the file: no judgments treaty, no forum-clause treaty, and no treaty channel for serving process or taking evidence either.
Conflicts & Capital, By the Numbers, Release 01
On the Radar
Oct 2026Switzerland: consultation closes on the ordinance implementing the Investment Screening Act.
Dec 2026AI Act Article 50(2) marking bites on systems already on the market, and the new Article 5 prohibitions take effect.
Dec 2026Revised Product Liability Directive, transposition deadline.
In this issue
Conflicts & Capital · September 2026
01
The Practitioner’s Chair Law stated as of Sept. 28, 2026
Virág Blazsek of the University of Leeds on how a government that rescues a bank or rewrites a securities law can change a cross-border deal that has already closed, and why deal documents cannot fully anticipate sovereign intervention.
The trigger looks through the buyer to whoever ultimately controls it. The standard for stopping a deal does not move, so what the widening delivers is a class of deals exposed on the one call-in clock the Regulation gives no end.
The clause the buyer negotiated cannot reach the claim that matters. The clause it acquired, sitting in the target’s end-user terms, is one a stranger can bring an action about, and the action reaches versions the target retired before the deal.
04
By the Numbers · Data Sources retrieved Aug. 20, 2026
116 entries, seven instruments. On the 115 states and territories, 98 percent are bound by the New York Convention against 28 percent by the 2019 Judgments Convention. The sharper number is 30: the entries bound by the New York Convention and by nothing else in the file.
05
By the Numbers · Data Sources retrieved Aug. 21, 2026
Fifty-four jurisdictions, one question: after completion, for how long can a screening authority still reach back and unwind or penalize the deal? Of the 44 that can, four give the buyer a date that holds. This edition adds what Article 4(4) does to each of the 27 Member States, row by row.
Foreign by Control: When the Buyer’s Passport Stops Mattering
From January 17, 2028 the EU screening trigger looks through the buyer to whoever ultimately controls it, so a European acquirer can be a foreign investor. The standard for stopping a deal does not move: that is still the Treaty, and Xella construes it strictly. What the widened trigger delivers is not more prohibitions. It is a larger class of deals inside the scope of a national authorization requirement, on the one clock the Regulation gives no end, under national nullity rules whose compatibility with Article 63 TFEU is contestable.
The clause the buyer negotiated cannot reach the claim that matters. The clause it acquired, sitting in the target’s end-user terms, is one a stranger can bring an action about, and the action reaches versions the target retired before the deal.
Foreign by Control: When the Buyer’s Passport Stops Mattering
The trigger moves from where the buyer is incorporated to who sits at the top of the control chain. The standard for stopping a deal does not move, so what the widening delivers is not more prohibitions. It is a larger class of deals inside a national authorization requirement, on the one clock the Regulation gives no end.
By Eric Martin, CICL Fellow, Center for International and Comparative Law
The EU’s new foreign-investment screeningA regime that reviews acquisitions for national-security or public-order risk and can block or condition them. regulation, Regulation (EU) 2026/1386, defines the investor it reaches by looking through the buyer to whoever ultimately controls it. It entered into force on July 16, 2026 and applies from January 17, 2028, the date by which every Member State must have established a screening mechanism and notified it to the Commission.1 From that date a deal run through a European holding company will not sit outside the Union regime merely because the acquirer is European on paper: the definition of a foreign investment reaches an investment carried out “through a foreign investor’s subsidiary in the Union,” which the Regulation defines as an undertaking established under the law of a Member State and directly or indirectly controlled by a foreign investor.6
And the word the whole trigger turns on is not defined. Article 2 runs to twenty-two definitions and not one of them is “control,” although at least four turn on it. Point (1) catches an investment enabling effective participation in the management or control of the target; point (5) confines “foreign investor” to a third-country national or an undertaking organized under the law of a third country; point (7) reaches that investor’s subsidiary in the Union where it is directly or indirectly controlled by it; and point (6) defines a beneficial owner by reference to who directly or indirectly owns or controls the investor or the target and on whose behalf control is exercised. Point (6) is the nearest thing to a benchmark the instrument contains, which makes it the first place an argument by analogy will start, and the analogy it invites is to anti-money-laundering practice rather than to merger control. Whether a third-country limited partner in a European fund controls that fund’s bolt-on acquisitions is not a question these definitions answer, and it is the question a fund lawyer will ask first.6
What widens. Screening stops being optional. Under Regulation (EU) 2019/452 Member States “may maintain, amend or adopt” a mechanism; under the new Regulation each “shall establish” one. And the sector floor becomes a Union floor rather than 27 national ones. Article 4(15) requires every Member State to impose a prior authorization requirement where the target develops, produces, or commercializes dual-use or defense-listed items, works on the semiconductor, quantum, or artificial-intelligence technologies listed in the Regulation’s own Annex I, is active in transport, energy, or digital infrastructure and is assessed as critical, handles listed strategic raw materials, is a named financial-market institution, or runs electoral systems.7
And the Regulation runs two clocks. For an investment that was not subject to a prior authorization requirement, the authority’s own-initiative power must last at least 15 months and may, at each Member State’s election, run to five years. For an investment that was subject to authorization and was never filed, or was filed only after completion, the floor is 24 months and the Regulation states no maximum, leaving the far end to the national mechanism each Member State must establish2A period after closing during which an authority can review and unwind a transaction it was never asked to clear..
The look-through is what decides which clock a deal is on. Take the buyer in Xella: a Hungarian company owned up the chain from Bermuda. Today that buyer is outside the scope of the Union regime, and the Court said so, holding that the assessment factor in Article 4(2)(a) of Regulation (EU) 2019/452 relates only to the ownership structure of a “foreign investor” as art. 2(2) defines it and does not extend the Regulation’s scope to a Member-State undertaking under third-country majority control. From January 17, 2028 that buyer is a foreign investor by definition. If the target sits in one of the Article 4(15) sectors, the deal is notifiable. And a deal team working from the map Regulation (EU) 2019/452 drew does not file it. That deal has not merely been added to the clearance map. The look-through makes it notifiable, the missed filing puts it on the Article 4(5) clock, which states no maximum, and the five-year ceiling in Article 4(4) was never available to it in any state of the world. Article 4(4) is confined to an investment not subject to a prior authorization requirement, so the moment the deal becomes notifiable it leaves that article behind, whether or not anyone files. What the diligence step decides is not which ceiling applies. It decides whether there is one.
What does not widen is the standard. The Regulation enlarges the set of transactions that must be reviewed. It does not touch the reason a deal can be blocked, and so it does not enlarge the set that can lawfully be stopped: that set is drawn by the Treaty and not by the trigger. Xella stated the standard six months before the Commission proposed this Regulation: an undertaking incorporated in a Member State is an EU company entitled to freedom of establishment whatever the origin of its shareholders, and grounds of public policy and public security, as derogations from a fundamental freedom, must be construed strictly; as to security of supply in particular, the Court held that such an objective may be relied on only where there is a genuine and sufficiently serious threat to a fundamental interest of society. The constraint is the Treaty, not the Regulation, and a regulation cannot authorize what the Treaty prohibits.
Which Treaty freedom is in issue decides how narrow the margin is, and the Regulation’s own recital 10 names both. Xella is an Article 49 holding: the Court said in terms that the case fell to be examined solely under freedom of establishment, because the acquirer was an EU company, and Article 52(1) excuses a restriction on that freedom only on grounds of public policy, public security, or public health. Where the investor is itself a third-country undertaking, which is the case the new Regulation is written for, the freedom is Article 63 TFEU, the derogation is Article 65(1)(b), and the Court has long held that capital movements to and from third countries take place in a different legal context. The margin in the Regulation’s central case is therefore wider than a reading of Xella alone would suggest.
Two different reviews also have to be kept apart. A challenge to the Regulation itself would be a review of a Union act adopted under Articles 114 and 207(2) TFEU, where the legislature has a wide margin and the question is proportionality to the objective pursued. A challenge to a prohibition decision taken by a Member State under it is a review of a national derogation from a fundamental freedom, where the margin is narrow and the burden sits on the Member State. Nothing said here is about the first; the whole of it is about the second.
What a regulation could do is harmonize the substantive test, so that a national measure fell to be judged against the harmonized standard rather than against the Treaty freedom directly. It does not do so in the provisions read for this piece. Nothing in Articles 1 to 5, and nothing in the recitals, tells a screening authority what it may prohibit. Articles 19 and 20, which govern assessment criteria and screening decisions, are not read here and are named in the margin as the place that question is answered. Recital 10 points the same way, requiring screening to comply with Articles 49 and 63 TFEU and carrying the Xella threshold across.8 A recital cannot create that constraint and is not being asked to; it is evidence that the legislature did not try to displace it.
A prohibition takes a decision. A nullity does not. That is where the widened trigger actually lands. Nothing here counts how often an authority blocks a deal or fines a buyer, and this column has no such data; the point is structural, which is that in most of these Member States the worst outcome does not wait for anyone to decide anything: an investment made without authorization is void in France, a nullity from the moment of the transaction in Latvia and from conclusion in Lithuania, and concluded under a suspensive condition by operation of law in Austria and Cyprus. Croatia is a fifth shape again: no fines at all, and an order to sell everything within nine months.Conflicts & Capital, By the Numbers, Release 02, rows for France, Latvia, Lithuania, Austria, Cyprus, and Croatia.
It is tempting to say that Xella’s threshold governs the veto and says nothing about the penalty for never asking. The nearest authority is twenty years old, and it is not on all fours. In Burtscher the Court held that Article 56(1) EC, now Article 63 TFEU, precludes national legislation “under which the mere fact that the requisite declaration of acquisition is submitted after the due date results in the retroactive invalidity of the property transaction concerned.” That was a prior-declaration regime for acquisitions, a filing made late, and automatic retroactive nullity, and it did not survive. The regimes listed above are triggered by the absence of any authorization, not by lateness, and a Member State will argue the distance: a formality penalty is one thing and evading a control system is another. And the Member State has an argument Burtscher never faced: the Vorarlberg regime was a national choice, while from January 17, 2028 the authorization requirement in the Article 4(15) sectors is a Union obligation, so what the sanction defends is the effectiveness of Union law. That does not exempt it from proportionality; it moves the case a long way from a land register. Three of the six impose the remedy the Court struck down in terms, France, Latvia, and Lithuania. The other three reach much the same practical place by a different route, a suspensive condition that never lets the transaction take effect, and whether Burtscher reaches that route is itself open. What the case gives a buyer is not an answer. It is that a court has already struck down retroactive invalidity imposed for the lapse of a deadline alone, which is enough to make the sanction contestable rather than settled.9
So both ends are constrained, at different intensities, and that is the useful way to hold this. A prohibition has to clear a demanding substantive test, and what a Member State must show is a genuine and sufficiently serious threat to a fundamental interest of society. A sanction for not filing has an easier justification available, because what is being defended is the effectiveness of a control system the Union now requires rather than a fundamental interest of society. But an easier justification is not an exemption from review, and proportionality is where the sanction is tested. Retroactive invalidity imposed for the lapse of a deadline alone is what Burtscher struck down. The exposure this piece is about is therefore not principally that the deal will be blocked. It is that an unfiled deal can sit for years inside a national nullity rule whose own compatibility with the Treaty is contestable, which is a worse position than either a clearance or a prohibition. On the clear cases the step that avoids it is a filing, and the memo that tells you to make one is cheap; on the arguable cases the memo ends where Article 2 does.
What actually happens is less dramatic than an unwinding and worse to live with. The Regulation says when an authority may open a review and not what it may do in the meantime, and the national instruments in Release 02 answer that. Cyprus bars the investor from exercising any rights in the target, voting and management included, until it complies. Croatia orders the shares sold within nine months, extendable by six, with the rights attaching to them suspended until the sale. Estonia empowers the authority to require transfer of the holding or reversal of the transaction. None of that is a prohibition, and none of it waits for a hearing on whether the deal threatened anything. It is the ordinary consequence of not having asked, and it lands on an asset the buyer is by then running.Conflicts & Capital, By the Numbers, Release 02, rows for Cyprus, Croatia, and Estonia.
The best objection to all of this sits in the same recital. Recital 10 makes “genuine and sufficiently serious threats to a fundamental interest of society” an example of public policy or public security rather than the test of it, and it then lists among the qualifying reasons a risk to “the supply of essential products or services.” That is the currency in which Hungary lost: the Court held that securing the construction sector’s supply of gravel, sand, and clay is not a fundamental interest of society. A Member State now has drafting to point at. It is answered on two grounds. A recital cannot widen a Treaty derogation that Article 52(1) TFEU and the case law construe strictly, and the same recital opens by requiring compliance with Articles 49 and 63 TFEU. And an example is not a substitute for the test: what qualifies as a fundamental interest is still for the Court. But the argument will be run, and it is where the first case under this Regulation is likely to start.
The harder version of the same objection is not about the recital’s wording at all. It is that the Regulation hands a Member State Union-sourced justification without ever writing a standard. Article 4(15) does not merely widen review, it compels a prior authorization requirement in named sectors, so a Member State imposing one is discharging a Union obligation rather than exercising a national choice. And recital 10 is not silent on the ground of decision: it lists the reasons of public policy and public security that qualify. Neither of those two carries a substantive standard on its own. A prior-authorization obligation fixes the trigger and not the outcome; a recital’s list is not a standard against which a prohibition is measured; and Article 4(16) leaves every Member State free to screen beyond the floor on grounds of its own. Whether the Regulation supplies a standard elsewhere is a question for Articles 19 and 20, which govern assessment criteria and screening decisions and are not read here, and the answer to it moves the margin in one direction or the other. What does not depend on it is the Treaty constraint itself: Articles 49 and 63 TFEU apply to a screening decision whatever the Regulation says about the grounds for taking one, and recital 10 says so.
Control, not the certificate of incorporation, now draws the clearance map. The consequence is not simply that more deals are on it. It is that a deal drawn onto the map by the look-through, and then left unfiled because the deal team used the old map, comes off the clock with a five-year ceiling and joins the clock that has none.
The deal has not merely been added to the clearance map. It has been put on the one clock the Regulation gives no end, by a diligence step nobody took.
The clock the filing decision picks
The Regulation runs two of them, and only one has an end.
Statutory floor Member State’s election, or unbounded Where the two part company
Both bars are drawn from Reg (EU) 2026/1386, arts. 4(4) and 4(5). Note which deal is which, because it is the opposite of what the numbers suggest: art. 4(4), with the five-year ceiling, governs the investment that never needed a filing, and art. 4(5), with no stated maximum, governs the one that needed a filing and did not make it. The bar with the higher floor is the bar with no roof. A third case is on neither bar: a deal subject to the requirement that was notified before completion and cleared leaves both articles behind, and what remains is whatever the national mechanism provides for reopening a clearance. Measured against what Member States have today, the ceiling in the first bar looks like a contraction for fifteen of the 27, and the One Number below says why it reaches far fewer and the floor in the second changes nothing for any of them, which is set out row by row in Release 02.
Also this month
Merger control
The $55 billion take-private of Electronic Arts by PIF, Silver Lake and Affinity Partners closed on August 4, 2026. The European Commission received the merger notification on June 16, 2026 (Case M.12213). Electronic Arts reported on July 30, 2026 that every regulatory approval required to complete the merger had been obtained, which is where the fact of clearance comes from here; the clearance decision itself was not on the Commission’s public case register when this was checked, and the specific date of July 23, 2026 rests on press reporting alone. The deal was also filed under the Foreign Subsidies Regulation.3 merger control and the subsidies regime each asked a question about the buyer’s state capital, whatever foreign-investment screening asked is not on any public record, and none of them addressed whose law governs a player-data claim after closing.
Foreign investment
Regulation (EU) 2026/1386 entered into force on July 16, 2026 and applies from January 17, 2028, repealing Regulation (EU) 2019/452.1 screening becomes mandatory in all 27 Member States, and the look-through to the acquirer’s ultimate control stops being an anti-abuse rule and becomes part of the definition. Under Regulation (EU) 2019/452 the EU-incorporated acquirer sat outside the Union regime’s scope, and the only Union route to it was the duty in art. 3(6) on Member States that already had a mechanism to prevent circumvention, which recital 10 confines to artificial arrangements that do not reflect economic reality. Xella says both halves: the scope of the 2019 Regulation was not extended to a Member-State undertaking under third-country majority control, and nothing in that file suggested circumvention.8
Enforcement
The 2019 Hague Judgments Convention entered into force for Albania and Montenegro on March 1, 2026 and for Andorra on June 1, 2026.4 the enforcement map is widening, but the United States signed in 2022 and has not ratified, so where the assets sit still decides whether a judgment is worth anything.
Technology and AI
The AI Act’s Article 50 transparency obligations began to apply on August 2, 2026, and on the same date the AI Act was added to Annex I of the Representative Actions Directive as point 68.5 an alleged AI Act breach is now a basis for a representative action where it harms the collective interests of consumers, and a qualified entity designated in advance in one Member State for cross-border actions can bring it before the courts of another. Providers whose systems were already placed on the market before August 2, 2026 have until December 2, 2026 to meet the Article 50(2) marking duty, under a four-month transition inserted into the AI Act by the Digital Omnibus.
One Number
24 months, no maximum
The minimum period Article 4(5) requires a Member State to give its screening authority over an investment that was subject to a prior authorization requirement and was not filed, or was filed after completion. For that case the Regulation fixes no outer limit at all.
Source: Reg (EU) 2026/1386, art. 4(5), read in the text. All 27 Member States already meet or exceed the 24-month floor, so the provision adds nothing to any of them, and it takes nothing away either: there is no ceiling in it. The ceiling sits in Article 4(4), which is confined to an investment not subject to a prior authorization requirement. How many of the 27 that ceiling actually contracts is not resolved by this release. Sixteen carry no outer limit today, but the file does not classify each untimed power onto the Article 4(4) track, and on the cells as published only Bulgaria, Romania, and Sweden record a power that is not conditioned on a missing filing, with Lithuania’s reading either way. See By the Numbers, Release 02.
Eric Martin, Foreign by Control: When the Buyer’s Passport Stops Mattering, Conflicts & Capital (Aug. 20, 2026), https://conflictsandcapital.netlify.app/brief/foreign-by-control.
Regulation (EU) 2026/1386 of the European Parliament and of the Council of 17 June 2026 on the screening of foreign investments in the Union and repealing Regulation (EU) 2019/452, OJ L, 2026/1386, 26.6.2026. Legal basis: Articles 114 and 207(2) TFEU. On the dates: art. 3(1) requires each Member State to establish a screening mechanism, and art. 3(2) requires notification of the implementing measures to the Commission “by 17 January 2028”; recital 66 states that the Regulation “should start to apply 18 months from its date of entry into force” and that Regulation (EU) 2019/452 continues to apply to foreign direct investments undergoing screening on, or completed by, the date of application. EUR-Lex ↗↩
Reg (EU) 2026/1386, art. 4(4): screening authorities must be empowered to act on their own initiative, for at least 15 months and up to a maximum of five years after completion, over a foreign investment within the scope of the national mechanism and not subject to a prior authorization requirement, where they have grounds to consider it may affect security or public order. Art. 4(5): at least 24 months, with no stated maximum, where the investment was subject to a prior authorization requirement and was not filed, or was filed after completion. EUR-Lex ↗↩
Case M.12213, PIF / ELECTRONIC ARTS, prior notification received June 16, 2026, OJ C, 1.7.2026. Completion on August 4, 2026 per Electronic Arts Inc., Form 8-K filed August 4, 2026, and the parties’ announcements of the same date; the 8-K of July 30, 2026 records that all regulatory approvals required to complete the merger had been obtained. The Commission’s clearance decision was not on the public case register when this piece was checked. That clearance was given is taken from the Form 8-K of July 30, 2026 recording that all required regulatory approvals had been obtained; the date of July 23, 2026 rests on press reporting alone and has not been confirmed against the register. CFIUS does not publish its decisions and is not named in the filings. ↩
Convention of 2 July 2019 on the Recognition and Enforcement of Foreign Judgments in Civil or Commercial Matters. Entry into force per the HCCH status table for Convention No. 41; the United States signed on March 2, 2022 and has not ratified. HCCH status table ↗↩
Regulation (EU) 2024/1689 (Artificial Intelligence Act), art. 113 (Chapter IV applies from August 2, 2026) and art. 110, adding the Act to Annex I of Directive (EU) 2020/1828 as point 68. Directive (EU) 2020/1828, arts. 2(1), 4(3) and 6(1): a representative action requires harm to the collective interests of consumers, and only a qualified entity designated in advance in another Member State for cross-border actions may sue there. On the four-month transitional period for systems already placed on the market before August 2, 2026, art. 111(4) of Regulation (EU) 2024/1689 as inserted by Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026, explained at recital 38. EUR-Lex ↗↩
Reg (EU) 2026/1386, art. 2, point (1): a foreign investment is “an investment of any kind, carried out either by a foreign investor itself or through a foreign investor’s subsidiary in the Union, aiming to establish or to maintain lasting and direct links between the foreign investor and a Union target, to which the foreign investor makes capital available in order to carry out an economic activity in a Member State, enabling effective participation in the management or control of that Union target”; art. 2, point (7): a foreign investor’s subsidiary in the Union is “an undertaking which is established under the laws of a Member State and directly or indirectly controlled by a foreign investor”; art. 2, point (5): a foreign investor is “(a) a natural person who does not hold the nationality of a Member State; or (b) an undertaking or entity established or otherwise organised under the laws of a third country”. Article 2 was read in full and runs to twenty-two numbered points, in the Official Journal’s own order and spelling: “foreign investment”; “greenfield investment”; “internal restructuring”; “request for authorisation”; “foreign investor”; “beneficial owner”; “foreign investor’s subsidiary in the Union”; “opaque ownership structure”; “Union target”; “filing”; “host Member State”; “screening”; “screening mechanism”; “screening decision”; “screening authority”; “completion”; “notifying Member State”; “multi-country transaction”; “multi-country notification”; “mitigating measure”; “contact point”; and “stockpiling”. None of them is “control”. Point (6) defines a beneficial owner as, among others, a natural person “who, directly or indirectly, own or control a foreign investor or Union target” or “on whose behalf the control over that foreign investment is exercised”. EUR-Lex ↗↩
Reg (EU) 2019/452, art. 3(1): Member States “may maintain, amend or adopt mechanisms to screen foreign direct investments in their territory”. Reg (EU) 2026/1386, art. 3(1): “Each Member State shall establish a screening mechanism in accordance with this Regulation.” Art. 4(15) lists the cases in which a Member State must impose a prior authorization requirement, by reference to Annex I to Regulation (EU) 2021/821, the Annex to Directive 2009/43/EC, Annex I to Regulation (EU) 2026/1386 itself for semiconductor, quantum and artificial-intelligence technologies, an annex described here only by the words of the cross-reference, Annex I not having been read for this piece, a risk-based assessment of critical transport, energy and digital infrastructure, Section I of Annex I to Regulation (EU) 2024/1252, named financial-market infrastructures, and electoral systems. Art. 4(16) permits Member States to go beyond that list; art. 4(17) disapplies art. 4(15) to greenfield investments. EUR-Lex ↗↩
Case C-106/22, Xella Magyarország, ECLI:EU:C:2023:568, paras. 44 to 48 (an undertaking formed under the law of a Member State is an EU company for the purposes of freedom of establishment, and “it does not follow from any provision of EU law that the origin of the shareholders… affects the right of those companies to rely on freedom of establishment”), paras. 66 and 67 (grounds of public policy and public security are derogations from a fundamental freedom and must be interpreted strictly; and, as to an objective linked to security of supply specifically, that it “may be relied on only if there is a genuine and sufficiently serious threat to a fundamental interest of society”) and paras. 41 to 43 and 49 (the case falls to be examined solely under freedom of establishment, not under the free movement of capital), and paras. 37 to 39 (the assessment factor in art. 4(2)(a) relates only to the ownership structure of a “foreign investor” as defined in art. 2(2), so it “does not mean that the scope of that regulation, as defined in Article 1(1) thereof, is extended to include investments made by undertakings organised in accordance with the laws of a Member State over which an undertaking of a third country has majority control”; and nothing in the file suggested a decision taken to counter circumvention within art. 3(6), whose scope recital 10 clarifies as “investments from within the Union by means of artificial arrangements that do not reflect economic reality… where the investor is ultimately owned or controlled by a natural person or an undertaking of a third country”). Reg (EU) 2019/452, art. 3(6) (“Member States which have a screening mechanism in place shall maintain, amend or adopt measures necessary to identify and prevent circumvention of the screening mechanisms and screening decisions”) and recital 10. Reg (EU) 2026/1386, recital 10: screening “should comply with Union law, and in particular with Articles 49 and 63 TFEU”, and any restriction resulting from it “should be justified by reasons of public policy or public security, including genuine and sufficiently serious threats to a fundamental interest of society”. EUR-Lex ↗↩
Case C-213/04, Burtscher v. Stauderer, judgment of 1 December 2005, operative part: “Article 56(1) EC precludes the application of national legislation such as the Vorarlberg Land Transfer Law (Vorarlberger Grundverkehrsgesetz) of 23 September 1993, in its amended version, under which the mere fact that the requisite declaration of acquisition is submitted after the due date results in the retroactive invalidity of the property transaction concerned.” The regime in issue was a prior-declaration requirement for the acquisition of land, and the sanction was automatic retroactive nullity on a late filing, irrespective of the reason for the delay. Article 56(1) EC is now Article 63 TFEU. The case is on the sanction, not on the prohibition, which is why it and not Xella is the authority the text turns to. It is not on all fours: the declaration in Burtscher was filed late, and the regimes the text lists are triggered by the absence of any authorization. Nothing in this judgment decides that case, and the text does not say it does. EUR-Lex ↗↩
The clause the buyer negotiated cannot reach the claim that matters. The clause it acquired is one a stranger can bring an action about, and the action reaches versions the target retired before the deal.
By Eric Martin, CICL Fellow, Center for International and Comparative Law
A governing-law clause binds the parties who signed it, and Regulation (EC) No 864/2007 (Rome II) assigns an injured third party’s claim by its own cascade without reference to it. That much is settled, and on the facts a textbook would use the cascade returns the law the general rule would have returned anyway. On the facts of a software target it does not, for three reasons that are all ordinary. The argument of this piece is about the other clause. The target’s own end-user agreement is not a shield the buyer inherits: after VKI v. Amazon a term reciting the trader’s home law and stopping there is capable of being unfair, point (2) of Annex I to Directive (EU) 2020/1828 makes an unfair term the subject of a representative action, Article 2(1) of that Directive reaches infringements that ceased before the action was brought, and Article 8(3) lets a qualified entity seek an injunction without any consumer coming forward or any proof of loss. The objection to that is that an injunction is cheap, and the answer is that the order is not the amendment: Article 8(2) allows a finding that the practice is an infringement and an order to publish it, Article 13(3) makes the trader notify the affected consumers at its own expense, and Article 15 makes the final decision evidence in every follow-on national redress action against the same trader for the same practice. Redress itself is a narrower and more national question, and the piece says where the line falls. The piece traces the cascade and says what it is worth; marks the seam between Rome II and Rome I Article 6, which are routinely run together; corrects the common reading of the 1973 Hague Products Liability Convention, which does not apply between a supplier and the person he supplied; and prices the two instruments that decide the size of the exposure, the expiry regime in Directive (EU) 2024/2853 and the collective-redress route in Directive (EU) 2020/1828. It also says what governs the units already shipped, which on any deal closing before December 2026 is almost all of them: Directive 85/374/EEC, which runs its own ten years and which never says whether software is a product at all.
Applies from January 11, 2009Regulation (EC) No 864/2007 (Rome II), arts. 31 and 32, as construed in Homawoo
Binds 26 of 27 EU Member StatesRome II, art. 1(4) and recital 40
Article 5, the product-liability cascadeRome II, art. 5
1973 Hague Products Liability Convention binds seven EU Member StatesHCCH status table, Convention No. 22; saved by Rome II, art. 28(1)
A consumer’s protection floor is uneven, and cannot be chosen away where it appliesRome I, art. 6(2)
Software is a product for units placed on the market after December 9, 2026; earlier ones stay under Directive 85/374/EECDirective (EU) 2024/2853, arts. 2(1), 4(1) and 21(1)
Ten years to start a claim, twenty-five for slow-emerging injury, for units inside the new Directive; ten years from putting into circulation for everything before itDirective (EU) 2024/2853, art. 17; Directive 85/374/EEC, art. 11
A choice-of-law term that recites only the trader’s law can be unfair, so far as it misleadsVKI v. Amazon, Case C-191/15, paras. 66 to 71
A qualified entity can bring that as a representative action, with no consumer coming forward, over a practice that has already stoppedDirective (EU) 2020/1828, arts. 2(1) and 8(3), and Annex I, point (2)
A United States buyer acquires a European software company, and the purchase agreement chooses New York law. Two years after closing, a consumer in Germany is injured by the product and sues. The buyer reaches for the New York choice-of-law clause, on behalf of the target it now owns. It does not apply, and that is the least of it. Under Regulation (EC) No 864/2007 (Rome II)The EU regulation fixing the law applicable to non-contractual obligations, such as tort and product-liability claims., the law governing a non-contractual claim, including product liability, is fixed by the Regulation’s own rules, and the only contract that could displace them is one this claimant is a party to.1 Three further instruments then decide what the claim is worth, how long it can be brought, and who may bring it. The acquisition agreement reaches none of them.
Which body of conflicts law runs at all depends on where she sues. If she sues in a Member State, everything below is live. If the claim is brought against the United States parent in a United States court, none of Rome II applies and that court runs its own choice-of-law analysis. The New York clause is still irrelevant there, for a reason that has nothing to do with any of this: she is a stranger to the acquisition agreement in New York exactly as she is in Frankfurt, and a contract cannot choose the law of a claim by someone who did not sign it. Everything that follows assumes she sues in a Member State, which on these facts is the likely place, because that is where she is and where Brussels Ia gives her a court. Inside the Union the forum splits with the claim. Her contract claim against the target sits in Section 4 of Brussels Ia: Article 17 of Brussels Ia confines that section to matters relating to a consumer contract and to the other party to it, Article 18(1) gives her the courts for the place where she is domiciled, and Article 19 permits departure only by an agreement made after the dispute arose, or one that widens her choice, or one between parties domiciled in the same Member State when they contracted. Her tort claim runs on Article 7(2) instead, where a jurisdiction agreement falls to be tested under Article 25 and Article 19 does not reach it. So the end-user forum clause is worth very little against the contract claim and rather more against the tort claim, which is the opposite of what a reader would guess.Regulation (EU) No 1215/2012, arts. 7(2), 17, 18(1), 19 and 25.
Why the clause does not travel
A choice-of-law clauseA contract term in which the parties select the governing law for disputes between them. is an agreement between the people who signed it. The consumer in Germany signed nothing. Rome II binds every Member State except Denmark and applies to events giving rise to damage occurring after January 11, 2009; in Croatia it has applied since accession on July 1, 2013.2
Article 14 does allow a choice of law for a non-contractual obligation, but only by the parties to that obligation. The consumer is not a party to the acquisition agreement, so there is no Article 14 choice to test and Articles 4 and 5 apply of their own force. That is the first of two reasons the agreement never gets near this claim. The second is that the buyer is not on the list of people who can be sued at all, and that list is not in Rome II; it is in the product-liability Directive, and it is set out further down. Article 14 matters one deal over, where buyer and seller might purport to choose the law of the tort as between themselves. There it is confined three times: the choice “shall not prejudice the rights of third parties”; a choice made before the event is available only where all the parties are pursuing a commercial activity and the agreement was freely negotiated; and two non-derogability rules apply. Under Article 14(2), where all the elements relevant to the situation are located in one country, choosing another country’s law cannot displace that country’s rules that cannot be derogated from by agreement. Article 14(3) does the same work one level up: where all those elements are located in Member States, choosing the law of a non-Member State cannot displace provisions of Union law that cannot be derogated from.3
Where the Article 5 cascade stops converging
Article 5(1) applies without prejudice to Article 4(2), so a common habitual residence of claimant and defendant takes priority. Failing that, the cascade runs to the injured person’s habitual residence, then the place the product was acquired, then the place the damage occurred, and each limb applies only if the product was marketed in that country. Within Article 5(1) itself a foreseeability override then applies: if the person claimed to be liable could not reasonably foresee the marketing of that product, or a product of the same type, in the country the cascade selects, the applicable law is instead that of the country where that person is habitually resident.4
On the facts above, every limb returns German law, and so does the general rule in Article 4(1). That convergence is what makes the cascade look skippable. It stops being skippable in three situations, and each of them is ordinary for a software target.
Where the product was not marketed in the country a limb selects, that limb fails and the analysis drops to the next one, so a target that sells into some Member States and not others has one answer per user, not one answer per product.
Where the seller could not reasonably foresee marketing in the country the cascade lands on, the override sends the claim to the law of the habitual residence of the person claimed to be liable, which will normally still be the target as producer rather than the buyer’s group, unless the buyer has actually moved the producing entity. And Article 4(2) displaces the whole cascade where claimant and defendant are habitually resident in the same country at the time the damage occurs. That looks like a planning lever and is not one, for a reason that sits in the second half of the provision people stop reading. Article 23(1) fixes a company’s habitual residence at its place of central administration rather than its place of incorporation. But the same paragraph goes on: where the event giving rise to the damage occurs, or the damage arises, “in the course of operation of a branch, agency or any other establishment,” the place of that establishment is treated as the habitual residence. So a common residence with a German consumer does not require moving central administration to Germany. A German development or support establishment in the chain that produced the damage will do it, and a buyer can trip into Article 4(2) by standing one up after closing without anybody deciding anything. Read Article 4(2) as a rule post-closing integration can walk into, not as one a buyer would ever choose.Rome II, art. 23(1), both subparagraphs.
At the end of the cascade sits the Article 5(2) escape, which applies the law of another country where the tort is manifestly more closely connected with it, and which expressly contemplates that such a connection may rest on a pre-existing relationship between the parties, such as a contract, closely connected with the tort. That is the clause a buyer should read twice, and it points away from the acquisition agreement entirely, at a document neither party to the deal negotiated and neither one drafted for this purpose.
Identifying the applicable law does not end the analysis. Two provisions let the forum’s own law back in whatever the cascade selected: Article 16 preserves the overriding mandatory provisions of the forum, and Article 26 permits refusal of a provision of the applicable law that is manifestly incompatible with the forum’s public policy. A third lets in a law that may be neither. Article 17 of Rome II requires the court, in assessing the conduct of the person claimed to be liable, to take account “as a matter of fact and in so far as is appropriate” of the rules of safety and conduct in force at the place and time of the event giving rise to the liability, which in a product case is usually where the thing was made or put into circulation and not where the user was hurt. Read that carefully, because it points the other way from the rest of this piece: that provision is as often the producer’s as the claimant’s, and it is how a target that built to one country’s standard gets to say so in a court applying another country’s law.Rome II, arts. 16, 17 and 26.
One caution before any of this runs. Where claimant and defendant are already in contract, which is the ordinary software case, the boundary between contract and tort is not drawn by which document a claimant produces. The Court of Justice draws it by asking whether the contract has to be interpreted to decide the claim: in Brogsitter a claim pleaded in tort was treated as contractual because the conduct complained of could be regarded as a breach of the contract’s terms, judged by reference to its purpose, while in Wikingerhof a claim stayed in tort because interpreting the contract was not indispensable to deciding it.5 A defect claim for physical injury normally stays in tort, and Rome II Article 5 applies to it. A claim that the product did not do what the terms warranted usually does not, and lands under Rome I instead. The two Regulations are to be read consistently with each other and with Brussels Ia, so the same line governs the applicable law and the forum.
A buyer can trip into Article 4(2) by standing up a German establishment after closing, without anybody deciding anything.
The clause you bought, and what it is good for
Rome I Article 6(2) does not floor a Rome II tort. Article 6 of Rome I governs consumer contracts, so if the claim is a product-liability tort under Rome II Article 5, Article 6(2) has nothing to say about it. What the consumer-contract regime does instead is reach this dispute from three directions, none of which is the tort itself. It supplies the pre-existing relationship Article 5(2) contemplates, which is how the end-user agreement gets into the tort analysis at all and is the escape the conclusion of this piece turns on. It governs any parallel contractual claim the same user brings. And through Directive 93/13/EEC it supplies the standard against which the clause itself can be attacked. Three routes in, and not one of them makes a Rome I provision govern a Rome II claim.
The acquisition agreement cannot reach the consumer, because she is a stranger to it. The target’s own end-user terms are a different matter: a license or subscription agreement between the target and the injured user is exactly the kind of pre-existing relationship Article 5(2) contemplates. It is also governed by its own regime, and that regime is not permissive. Under Article 6(1) of Regulation (EC) No 593/2008 (Rome I), a contract between a consumer and a professional who directs activities to the consumer’s country is governed by the law of the consumer’s habitual residence. The parties may choose another law, but Article 6(2) provides that the choice “may not, however, have the result of depriving the consumer of the protection afforded to him by provisions that cannot be derogated from by agreement” under the law that would otherwise apply.6 A EULA choosing the target’s home law therefore sets a floor and not a ceiling: the chosen law keeps governing, and on any point where the consumer’s own law has a rule that cannot be derogated from and is more protective, that rule is added on top of it.
Three limits keep that from being a general rule, and each of them is a diligence question. Article 6(1) reaches only a contract concluded for a purpose outside the customer’s trade or profession, with a professional who pursues or directs activities to the country of her habitual residence, so a business user of the same software sits outside Article 6 altogether and falls back to Articles 3 and 4, where the clause does what it says. Article 6(4)(a) takes out a contract of services to be supplied exclusively in a country other than the consumer’s habitual residence. And the comparison runs point by point rather than regime by regime, so she does not import her own law wholesale; she gains only the rules of it that cannot be derogated from and that are more protective on the point actually in issue. The floor is real and it is uneven.6
And the clause can be unfair merely for saying so. In VKI v. Amazon the Court of Justice held that a term in a trader’s general conditions, not individually negotiated, providing that a contract concluded in the course of electronic commerce is governed by the law of the trader’s own Member State “is unfair in so far as it leads the consumer into error by giving him the impression that only the law of that Member State applies to the contract, without informing him that under Article 6(2) of Regulation No 593/2008 he also enjoys the protection of the mandatory provisions of the law that would be applicable in the absence of that term.” The Court was careful, and the care is the part a summary drops. The test it applied is the ordinary one in Article 3(1) of Directive 93/13/EEC, so the term has to cause a significant imbalance in the parties’ rights and obligations contrary to good faith; the Court said such a term is unfair “only in so far as it displays certain specific characteristics inherent in its wording or context” that do that. Choice-of-law terms are not unfair as such, the holding is confined to the extent (“in so far as”) the term misleads, and whether a particular clause misleads is for the national court on all the circumstances.7 So the finding is available where a clause recites the chosen law and stops there, which says the first half of the position and not the second. It is not available against a clause that carries the second half. A sentence saying that mandatory consumer protections of the user’s own country continue to apply is outside what the Court held unfair, and drafting it in is the cheapest thing on any list in this piece. What it does not do is retire the versions already shipped.
The same judgment carries a second holding that belongs in a conflicts column. The law applicable to an action for an injunction against the use of unfair terms is determined under Article 6(1) of Rome II, as unfair competition affecting the collective interests of consumers, while the law applicable to the assessment of any particular term is always determined under Rome I. One dispute, two Regulations, and the answers can differ.7
What survives Rome II
Rome II does not occupy the whole field. Article 28(1) saves pre-existing conventions laying down conflict-of-law rules for non-contractual obligations to which one or more Member States were party when Rome II was adopted. Article 28(2) then claws back, as between Member States, those conventions concluded exclusively between Member States. Because the 1973 Hague Products Liability Convention also binds states outside the Union, namely Norway, Serbia, North Macedonia, and Montenegro, Article 28(2) does not reach it, and it continues to govern in the seven Member States bound by it: France, Spain, Luxembourg, the Netherlands, Finland, Croatia, and Slovenia.8
But it does not govern the case in the opening paragraph, and the reason is not in its connecting factors. The second paragraph of Article 1 provides that where the property in, or the right to use, the product was transferred to the person suffering damage by the person claimed to be liable, the Convention does not apply to their liability inter se. In the ordinary software case, where the user downloads or subscribes directly from the target, the Convention is excluded as between exactly those two, whichever contracting state she sues in. A French court would apply Article 5 of Rome II to that claim, not the Convention.
Where the Convention does bite is the case an acquisition actually creates: a chain with more than two links. The claim against the component supplier, the upstream manufacturer, or the group company that did not itself supply the user is not caught by the inter se exclusion, and in a contracting state the Convention governs it. The forum decides whether the Convention is in play at all; the pairing of this claimant with this defendant decides whether the claim escapes the inter se exclusion. Both have to be checked. It binds the courts of contracting states, applies irrespective of the nature of the proceedings, and applies even where it points to the law of a non-contracting state. Its connecting factors are also different from Rome II’s: Article 4 requires the place of injury to coincide with the claimant’s habitual residence, the defendant’s principal place of business, or the place of acquisition; Article 5 prefers the claimant’s habitual residence on a similar coincidence; Article 6 falls back to the defendant’s principal place of business unless the claimant bases the claim on the law of the place of injury, which is the one point in the cascade where the claimant chooses; and Article 7 carries a foreseeability defense of its own. Two claims out of one accident, litigated in the same court, can therefore run on different instruments.9
One further caution before anyone prices the Convention. Its definition of a product, “natural and industrial products, whether raw or manufactured and whether movable or immovable,” sits awkwardly with software, and unlike Union law it has not been revisited: the definition in Article 2(a) is the one concluded on October 2, 1973, and the Convention has no equivalent of the definition in Directive (EU) 2024/2853.9
The clock the buyer has to price
Choice of law decides which rules apply. Limitation decides whether the claim exists at all, and here Union law has just moved. Directive (EU) 2024/2853 applies to products placed on the market or put into service after December 9, 2026, which is also its transposition deadline, and it settles the question the 1973 Convention leaves open: “product” expressly “includes electricity, digital manufacturing files, raw materials and software.” Free and open-source software developed or supplied outside a commercial activity is carved out.10
Which means that on a deal closing now, almost none of the installed base is in that Directive at all. Everything placed on the market before December 9, 2026 stays under Directive 85/374/EEC, and the old Directive does not answer the question the new one settles in a clause. Article 2 defines a product as “all movables even if incorporated into another movable or into an immovable” and then adds one inclusion, that “‘product’ includes electricity.” It names electricity and not software, and whether software is a movable for that purpose is argued, not decided, and is argued in national transposing law. That is the live question on a target’s existing releases, and until December 2026 it is worth more to a buyer than anything in the new Article 17, because it decides whether the pre-2027 base carries strict liability under a Union regime at all. The old Directive’s own outer period has the same shape as the new one: ten years from the date the producer put the actual product that caused the damage into circulation, unless the injured person has instituted proceedings in the meantime. So the schedule this piece asks for later is the right request either way; only the instrument it feeds changes.Council Directive 85/374/EEC, arts. 2 (as amended by Directive 1999/34/EC), 10 and 11.
Three of its provisions decide the shape of the buyer’s exposure. None of them can be bargained with the injured person. All of them can be allocated between buyer and seller, which is a different thing and comes later.
Liability cannot be limited against the injured person. Article 15 requires Member States to ensure that an economic operator’s liability “is not, in relation to the injured person, limited or excluded by a contractual provision or by national law.” The limitation of liability in the target’s terms is not a defense to her claim, whatever it is worth between the parties who agreed it.
The outer limit is ten years, or twenty-five. Article 16 of the Directive sets a three-year limitation period, running from the day the injured person knew or should have known of the damage, the defectiveness, and the identity of the economic operator liable under Article 8. Article 17 of the Directive sets the outer periods, and frames them as the extinction of the injured person’s entitlement rather than as a bar on proceedings: ten years from the placing on the market or putting into service under art. 17(1)(a), and twenty-five under art. 17(2) where she could not bring proceedings within the ten because of the latency of a personal injury. Both are subject to a carve-out worth reading, because it changes what a buyer is pricing: the entitlement expires “unless that injured person has, in the meantime, initiated proceedings against an economic operator.” The period is a deadline for starting a claim, not a date on which a live one dies. Note what each period bounds. The Directive’s Article 17 extinguishes the injured person’s entitlement against the economic operator; a contractual survival period bounds only the buyer’s recourse against the seller, so the two do not net against each other, they sit one behind the other. The ten years also runs unit by unit from the day each product was placed on the market, so the tail on an installed base is shorter than ten years from closing, while every release the buyer ships afterwards starts a fresh ten in the buyer’s own name. And the Directive’s three years from knowledge is the period that usually operates; its Article 17 is the backstop that decides whether the claim can exist at all.
A substantial modification moves the start date, and the bar is higher than it sounds. For a substantially modified product the ten years runs instead from the date that product was made available following the modification. What counts is defined, and it is not a version number: the modification must be treated as substantial by product-safety rules, or else must both change the product’s performance, purpose or type outside the manufacturer’s own initial risk assessment and change the nature of the hazard, create a new one, or raise the level of risk. A release inside the roadmap does not clear that. A re-architecture that introduces a hazard the original assessment never contemplated may. And the deeming provision that turns a modifier into a manufacturer is conditioned on the modification being made outside the manufacturer’s control, so a buyer shipping releases through the target it now owns is answering in its own name rather than reviving anything of the seller’s.
The tail is a schedule, not a period
Where the ten years actually runs, for four units of the same product. The earliest unit here ships six years before closing, so every unit sits inside Directive (EU) 2024/2853 only if closing falls after December 9, 2032. No deal closing this decade looks like that: on a real installed base the bars fall on both sides of the line, and the ones on the near side are governed by Directive 85/374/EEC instead. The chart is drawn on one instrument to show the shape of the period; the caption says what the shape hides.
Ten years from that unit’s placing on the market A unit the buyer places, in its own name Contract survival, from closing
n = one illustrative product with four release dates; the release dates are chosen to show the shape, and a reader should substitute the target’s own placing-on-the-market schedule. Article 17(1)(a) of Directive (EU) 2024/2853 runs its ten years from the date the individual product was placed on the market or put into service, and art. 17(1)(b) restarts it from the date a substantially modified product was made available, so the exposure on an installed base is ten years less the age of that base, and every unit the buyer ships after closing starts a fresh ten in the buyer’s own name. The dashed lane is drawn short deliberately: a survival period bounds the buyer’s recourse against the seller, not the injured person’s entitlement, so the two lanes do not net against each other. Not drawn: the twenty-five years art. 17(2) allows where a personal injury emerges too slowly for the ten, which would run off the right edge for every bar here; nor the carve-out in art. 17(1) itself, under which the entitlement survives the period where the injured person has initiated proceedings in the meantime, so a bar is the last date on which a claim can be started and not the date on which a pending one dies. And one thing the shape hides: for a deal closing in the next few years the installed base straddles December 9, 2026, and the units on the wrong side of that date are not on this chart at all. They stay under Directive 85/374/EEC, which runs its own ten years from the putting into circulation of the individual product and carries neither the disclosure duty in art. 9 nor the presumptions in art. 10. The diligence request has to ask which side of that date each release falls on, because the answer changes the instrument and not only the number.
The evidential position moves in the same direction. Article 9 requires disclosure of relevant evidence at the claimant’s request once the claim is shown to be plausible, and Article 10 presumes defectiveness where the defendant fails to disclose, where the product breaches mandatory safety requirements, or where an obvious malfunction occurred in reasonably foreseeable use, and presumes causation where the damage is of a kind typically consistent with the defect. It also lets the court presume either, or both, where technical complexity makes proof excessively difficult and the claimant shows the proposition is likely. The presumptions are rebuttable, which is not the same as neutral: each of them shifts the burden to the defendant at precisely the point where the claim would otherwise fail for want of proof.10
Whether the claim reaches the buyer at all
Everything above assumes a defendant. Article 8 says who that is, and the answer is narrower than a deal team expects: the manufacturer of the defective product, the manufacturer of a defective component integrated within the manufacturer’s control, and, where the manufacturer sits outside the Union, the importer, the authorized representative, and failing both the fulfilment service provider, with the distributor picked up under Article 8(3) of the product-liability Directive only where the injured person asks it to name an economic operator upstream and it fails to do so within one month of the request. Note that the first limb carries no requirement that the manufacturer be established in the Union; the Union-establishment condition attaches to limb (c), which adds the importer and the others alongside the manufacturer rather than in place of it. An acquiring parent that neither makes the product nor places it on the Union market is on none of these limbs, and so is not itself a defendant. What it bought is a subsidiary that is one, and every euro of that liability still runs through the equity it paid for. Whether it stays off the list is the next question.
Which makes post-closing integration a liability decision, and it is not usually taken as one. Move development out of the Union and the offshore group company becomes the manufacturer under Article 8(1)(a), which carries no Union-establishment requirement at all, while the Union entity that places the product on the market becomes the importer under Article 8(1)(c)(i). Article 8(1)(c) adds the importer “without prejudice to the liability of that manufacturer,” so this is not a relocation of exposure onto the subsidiary. It is an addition, and the larger half of it lands on the group company that now does the developing. Ship a release that clears the Article 4, point (18) threshold from a company outside the original manufacturer’s control, and Article 8(2) makes that company a manufacturer of the modified product in its own name.
One limb does not work this way, and the difference is worth keeping straight. An authorized representative is defined as a person “who has received a written mandate from a manufacturer,” so nobody becomes one by shipping. It is a box an integration plan ticks deliberately or not at all, which makes it the only limb nobody enters by accident. The rest happen whether or not anyone notices, and that is the point: what the buyer does after closing decides whether it is a defendant, and post-closing integration is the only part of this analysis the buyer controls.Directive (EU) 2024/2853, art. 8(1) to (3); art. 4, points (10), (11), (12) and (18).
And who gets to bring it
Here is where the clause the buyer acquired stops being a document and becomes a claim, and it is worth being exact about which kind. After VKI, a term reciting the trader’s own law and saying nothing about Article 6(2) is capable of being unfair. Point (2) of Annex I to Directive (EU) 2020/1828 makes Directive 93/13/EEC one of the instruments a representative action can be brought under, and Article 2(1) reaches infringements “where those infringements ceased before the representative action was brought.” What that buys is an injunction, and it buys it on easy terms: under Article 8(3) of the representative-actions Directive no individual consumer has to come forward to be represented, and the qualified entity need not prove loss to any of them or intent or negligence on the trader’s part. A term the target retired is therefore an exposure that survives the buyer’s arrival.
How far back that reaches is worth stating exactly, because it is stated loosely almost everywhere. Article 22 fixes which regime runs, not how old the practice may be. The transposing provisions govern representative actions brought on or after June 25, 2023, and actions brought before that date run on Directive 2009/22/EC instead. The only thing Article 22 dates at the other end is the suspension and interruption of limitation periods under Article 16, which it confines to redress claims based on infringements occurring on or after June 25, 2023. Nothing in it puts a floor under how old an infringement can be for the injunctive limb. That is a national limitation question, and it is one to put to local counsel rather than to answer out of the Directive.
The obvious objection is that an injunction is cheap. A partner will say the remedy is an order to change a clause, that changing a clause is a paralegal afternoon, and that nobody prices a paralegal afternoon. That is right about the clause and wrong about the order, and three provisions of the same Directive say why. A definitive injunctive measure may include a measure establishing that the practice constitutes an infringement, and an obligation to publish the decision or a corrective statement, in each case “if provided for in national law.” Article 13(3) then requires the court to make the trader inform the consumers concerned of the final decision, at the trader’s expense and individually where appropriate, unless they are informed some other way. And Article 15 makes a final decision of any Member State’s court on the existence of an infringement usable by all parties as evidence in any other action for redress against the same trader for the same practice, in accordance with national law on evaluation of evidence. The amendment is the afternoon. The exposure is a finding, delivered to the affected users at the buyer’s own cost, that every follow-on national claim can put into evidence. Each of those three is conditioned on national law, which is what makes this a diligence request rather than a number.
Redress is a different question, and this is where a client alert will overreach. Article 2(2) provides that the Directive “does not affect rules under Union or national law establishing contractual and non-contractual remedies”, and Article 9(1) gives remedies only “as appropriate and as available under Union or national law”. The Directive is a procedural conduit; it creates no cause of action. What Directive 93/13 supplies for an unfair term is that the term does not bind, which for a choice-of-law recital returns the consumer to the Article 6(2) floor she had anyway. And Article 9(3) requires consumers not habitually resident in the forum Member State to opt in expressly before a redress measure binds them. So the honest statement of the exposure is an injunction across the installed base of terms, the finding that can travel with it, and a money claim only where national law supplies the cause of action, rather than one collective judgment on behalf of everybody who ever clicked accept.11
The last instrument decides whether the claim arrives one user at a time. Under Directive (EU) 2020/1828, Member States must ensure that qualified entities “designated in advance in another Member State for the purpose of bringing cross-border representative actions can bring such representative actions before their courts or administrative authorities.” The Directive applies to domestic and cross-border infringements alike, and expressly “where those infringements ceased before the representative action was brought,” so a practice the target stopped at closing is still actionable afterwards. Annex I lists, at point (2), Directive 93/13/EEC on unfair terms, which is the limb that reaches the end-user agreement discussed above, and, at point (1), Directive 85/374/EEC on liability for defective products, which continues to apply to products placed on the market before December 9, 2026 and whose references are construed as references to the new Directive for products placed after it.11
Put them together and the answer to the question the buyer asked is not the interesting part. The New York clause never governed this claim and nobody competent thought it did. A German consumer’s tort claim is governed by German law under Rome II whatever New York law says, subject to one escape, Article 5(2), which can carry the tort to the law of another country where the end-user relationship makes the connection with that country manifestly closer. Read the provision precisely: it selects a country’s law, not whichever law the end-user agreement chose. A New York clause in a EULA is not by itself a connection with New York, and the escape more often lands on the law of the place the relationship actually sits. The floor of her contractual protection is set by Rome I Article 6(2) whatever the end-user agreement says, and the clause that told her otherwise may itself be unfair. The outer limit on the claim is ten years, and which instrument supplies it turns on one date: units placed on the market after December 9, 2026 run on Article 17 of Directive (EU) 2024/2853, which also allows twenty-five for a slow-emerging injury and restarts the ten on a substantial modification, and everything shipped before then runs on Article 11 of Directive 85/374/EEC, which does neither. And a qualified entity in another Member State can bring the whole thing collectively, about conduct that stopped before the buyer arrived.
The interesting part is the direction the exposure runs. Everything the buyer negotiated points one way, at the seller, on a period the parties set. Everything that decides the size of the claim points the other way, at the target’s own users, on periods the parties cannot set, brought under a clause the target wrote for a different purpose entirely. A deal team that reads the end-user agreement as boilerplate has read the only document in the data room that a stranger can sue on.
Insurance does not close this
The reasonable objection to everything above is that a buyer does not carry this risk on the contract at all. It takes a product-safety and compliance representation, backs the representation with warranty and indemnity insurance, and moves on. Three features of the new regime cut against that answer, and none of them can be underwritten away.
First, a warranty and indemnity policy responds to the breach of a representation, tested as at closing. Neither outer period describes a state of affairs at closing. Take the new Directive, which is where the tail is headed and where every unit shipped from December 9, 2026 lands: its Article 17 does not describe a state of affairs at closing; it describes an entitlement that arises when someone is injured, which may be years afterwards and on a unit shipped years before. A representation that the products complied with applicable law when the deal signed can be perfectly true and leave the claim untouched. Second, Article 15 makes the target’s own limitation of liability ineffective against the injured person, so whatever the underwriter priced against on the target’s side of the ledger is not there. And third, because that Article runs ten years from the placing on the market of each unit, the tail is not a single period a policy can be written to: it is a schedule of overlapping periods, one per release, and the buyer’s own post-closing releases start new ones in the buyer’s own name.
The practical consequence is a gap that can be computed rather than argued about. Ask for the placing-on-the-market schedule by release, run the ten years from each, and set the resulting curve against the policy period and against the survival period. Where the curve runs past both, someone is carrying it, and the memo should say who. And keep the two policies apart, because Rome II Article 18 reaches only one of them. It permits a direct action against “the insurer of the person liable to provide compensation” where either the law applicable to the non-contractual obligation or the law applicable to the insurance contract so provides. A warranty and indemnity insurer is not that person’s insurer; it insures the buyer’s claim against the seller, and the injured consumer is nowhere in that contract. The policy Article 18 is about is the target’s own product liability tower, which is a different diligence request, a different limit, and usually a different broker.Directive (EU) 2024/2853, arts. 15, 16 and 17; Rome II, art. 18.
How to draft around what you cannot choose
None of these six moves is the indemnity, which is where most of this conversation starts and ends.
Ask for every version of the end-user terms in force since June 25, 2023, not just the current one. Article 2(1) of Directive (EU) 2020/1828 reaches infringements that ceased before the action was brought, and after VKI the infringement can be a choice-of-law recital that told the user only half the position. June 25, 2023 is the date the new regime took over from Directive 2009/22/EC for actions brought after it, not a floor on how old the practice can be, so ask for the library going back as far as the target kept one and let local counsel say what its own limitation law does with the early versions. Nobody asks for a retired clause library in diligence, and it is one email.
Compute the tail rather than assuming it. The remaining exposure on the installed base is ten years less the age of that base, which is a number, and it is usually shorter and more negotiable than the headline suggests. Ask which releases fall before December 9, 2026 while you are at it: those stay under Directive 85/374/EEC, on a different evidential footing, and the schedule that answers the first question answers this one too.
Read the end-user agreement as a conflicts instrument and not as boilerplate. Check whether its choice-of-law clause tells the user about Article 6(2) or only about the trader’s law, because that is the difference VKI turns on, and check the forum clause on both of the lives it leads. Nobody drafts a EULA expecting it to be read as evidence of a pre-existing relationship under Rome II, which is the use this piece is about.
Decide who carries the pre-closing tail, and prefer cover to a promise. A ten-year indemnity from a fund that dissolves in year two is a paper asset. A contingent-liability or known-issue policy survives the seller; an indemnity does not. If the seller will not fund one, the buyer is carrying it, and one sentence in the memo should say so.
Treat the integration plan as a liability decision. Where development sits, which entity ships, and whether a release clears the substantial-modification threshold each decide who is on the Article 8 list. It is the part of this analysis the buyer controls, and it is usually settled by people who have never read Article 8 of the Directive.
Map the Member States where the product is sold and where the components come from. Seven of them run a different instrument, and it changes the answer only for the claims the target did not itself supply, which is exactly the set an acquisition creates.
Eric Martin, What Rome II Does to Your Governing-Law Clause, Conflicts & Capital (Aug. 20, 2026), https://conflictsandcapital.netlify.app/cornerstone/rome-ii-governing-law.
Regulation (EC) No 864/2007 of the European Parliament and of the Council of 11 July 2007 on the law applicable to non-contractual obligations (Rome II), 2007 O.J. (L 199) 40. EUR-Lex ↗↩
Rome II, arts. 31 and 32, read with Case C-412/10, Homawoo v. GMF Assurances SA, ECLI:EU:C:2011:747 (Nov. 17, 2011), holding that the Regulation applies only to events giving rise to damage occurring after January 11, 2009. Croatia acceded to the Union on July 1, 2013. ↩
Rome II, art. 14(1) (the choice “shall not prejudice the rights of third parties”), art. 14(1)(b) (pre-event choice only where all the parties pursue a commercial activity, by an agreement freely negotiated), art. 14(2) and art. 14(3) (non-derogable rules of the country, and of the Union, where all relevant elements are located there). ↩
Rome II, art. 5(1)(a)–(c) (the cascade, each limb conditional on the product having been marketed in that country), the closing subparagraph of art. 5(1) (foreseeability override, applying the law of the habitual residence of the person claimed to be liable), art. 4(2) (common habitual residence takes priority) and art. 5(2) (manifestly closer connection, which “might be based in particular on a pre-existing relationship between the parties, such as a contract”). ↩
On the contract and tort boundary, Case C-548/12, Brogsitter v. Fabrication de Montres Normandes, ECLI:EU:C:2014:148 (Mar. 13, 2014), and Case C-59/19, Wikingerhof v. Booking.com, ECLI:EU:C:2020:950 (Nov. 24, 2020). Both are decided on the jurisdictional heads of Regulation (EU) No 1215/2012 and its predecessor rather than on Rome I or Rome II; recital 7 of each of Regulation (EC) No 593/2008 and Regulation (EC) No 864/2007 requires the substantive scope of those Regulations to be consistent with that Regulation, which is why the same line is used here. EUR-Lex ↗↩
Regulation (EC) No 593/2008 (Rome I), art. 6(1) (a consumer contract is governed by the law of the consumer’s habitual residence where the professional pursues or directs activities to that country and the contract falls within the scope of those activities) and art. 6(2): a choice of law “may not, however, have the result of depriving the consumer of the protection afforded to him by provisions that cannot be derogated from by agreement by virtue of the law which, in the absence of choice, would have been applicable on the basis of paragraph 1”. Note the carve-outs in art. 6(4), of which art. 6(4)(a), services supplied to the consumer exclusively in a country other than that of his habitual residence, is the one a software supplier should check. Compare arts. 3(3) and 3(4), the general non-derogability provisions. EUR-Lex ↗↩
Case C-191/15, Verein für Konsumenteninformation v. Amazon EU Sàrl, ECLI:EU:C:2016:612 (July 28, 2016), operative points 1 and 2. On the term: unfair under art. 3(1) of Directive 93/13/EEC “in so far as it leads the consumer into error by giving him the impression that only the law of that Member State applies to the contract, without informing him that under Article 6(2) of Regulation No 593/2008 he also enjoys the protection of the mandatory provisions of the law that would be applicable in the absence of that term, this being for the national court to ascertain in the light of all the relevant circumstances”; see paras. 66 to 71, and para. 67 (such a term is unfair “only in so far as it displays certain specific characteristics inherent in its wording or context which cause a significant imbalance in the rights and obligations of the parties”). On the split: the law applicable to an action for an injunction is determined under art. 6(1) of Rome II, that applicable to the assessment of a particular term always under Rome I, both without prejudice to art. 1(3) of each Regulation; see paras. 42, 43 and 49. EUR-Lex ↗↩
Convention of 2 October 1973 on the Law Applicable to Products Liability, saved by Rome II, art. 28(1). In force for France, Spain, Luxembourg, the Netherlands, Finland, Croatia, and Slovenia, and for Norway, Serbia, North Macedonia, and Montenegro outside the Union. Belgium, Italy, and Portugal signed but never ratified. HCCH status table ↗↩
Convention of 2 October 1973, art. 1, second paragraph: “Where the property in, or the right to use, the product was transferred to the person suffering damage by the person claimed to be liable, the Convention shall not apply to their liability inter se.” Art. 1, third paragraph (application irrespective of the nature of the proceedings); art. 2(a) (“the word ‘product’ shall include natural and industrial products, whether raw or manufactured and whether movable or immovable”); arts. 4, 5 and 6 (the connecting factors, each requiring a coincidence between two of the place of injury, the claimant’s habitual residence, the defendant’s principal place of business and the place of acquisition, failing which the defendant’s principal place of business applies unless the claimant bases the claim on the law of the place of injury); art. 7 (foreseeability defense); art. 11 (application independent of reciprocity). HCCH ↗↩
Directive (EU) 2024/2853 on liability for defective products, art. 2(1) (applies to products placed on the market or put into service after December 9, 2026), art. 2(2) (free and open-source software developed or supplied outside the course of a commercial activity is excluded), art. 4(1) (“‘product’ means all movables, even if integrated into, or inter-connected with, another movable or an immovable; it includes electricity, digital manufacturing files, raw materials and software”), art. 4, point (18) (definition of “substantial modification”: substantial under Union or national product-safety rules, or, where those lay down no threshold, a modification that both changes the product’s original performance, purpose or type without that change having been foreseen in the manufacturer’s initial risk assessment and changes the nature of the hazard, creates a new hazard or increases the level of risk), art. 8(2) (a person who substantially modifies a product “outside the manufacturer’s control” and thereafter makes it available is considered its manufacturer), art. 9 (disclosure of evidence), art. 10 (burden of proof and the rebuttable presumptions), art. 15 (liability “is not, in relation to the injured person, limited or excluded by a contractual provision or by national law”), art. 16 (a three-year limitation period running from knowledge of the damage, the defectiveness and the identity of the operator liable under art. 8; art. 16 contains neither outer period), art. 17 (the expiry periods: ten years running from the placing on the market or, for a substantially modified product, from the date it was made available following the modification, and twenty-five years where the injured person could not bring proceedings within ten because of the latency of a personal injury), art. 21 (Directive 85/374/EEC repealed with effect from December 9, 2026, continuing to apply to products placed on the market or put into service before that date, and references to it construed as references to this Directive) and art. 22(1) (transposition by December 9, 2026). EUR-Lex ↗↩
Directive (EU) 2020/1828 on representative actions, art. 2(1) (scope, covering domestic and cross-border infringements “including where those infringements ceased before the representative action was brought or where those infringements ceased before the representative action was concluded”), art. 4(3) (criteria for designation as a qualified entity for cross-border actions), art. 6(1) (“Member States shall ensure that qualified entities designated in advance in another Member State for the purpose of bringing cross-border representative actions can bring such representative actions before their courts or administrative authorities”) and Annex I, points (1) and (2), listing Directive 85/374/EEC and Directive 93/13/EEC respectively. On what a qualified entity can actually obtain: art. 7(4) (a qualified entity must be entitled to seek “at least… (a) injunctive measures; (b) redress measures”); art. 8(3) (“In order for a qualified entity to seek an injunctive measure, individual consumers shall not be required to express their wish to be represented by that qualified entity”); art. 2(2) (the Directive “does not affect rules under Union or national law establishing contractual and non-contractual remedies”); art. 9(1) (a redress measure requires remedies “as appropriate and as available under Union or national law”); and art. 9(3) (consumers not habitually resident in the Member State of the forum “have to explicitly express their wish to be represented” in order to be bound). On what an injunction is worth: art. 8(2) (a definitive measure “may include, if provided for in national law: (a) a measure establishing that the practice constitutes an infringement…; and (b) an obligation to publish the decision on the measure in full or in part… or an obligation to publish a corrective statement”); art. 8(3) (no consumer need express a wish to be represented, and the qualified entity need not prove “actual loss or damage on the part of the individual consumers affected” or “intent or negligence on the part of the trader”); art. 13(3) (the court or administrative authority “shall require the trader to inform the consumers concerned by the representative action, at the trader’s expense, of any final decisions providing for the measures referred to in Article 7”, individually where appropriate, the obligation not applying where those consumers are informed in another manner, and Member States being free to require the qualified entity to ask for it); and art. 15, “Effects of final decisions” (a final decision of a court or administrative authority of any Member State on the existence of an infringement harming the collective interests of consumers “can be used by all parties as evidence in the context of any other action before their national courts or administrative authorities to seek redress measures against the same trader for the same practice, in accordance with national law on evaluation of evidence”). On reach in time: art. 22(1) (the transposing provisions apply to representative actions brought on or after June 25, 2023, and the provisions transposing Directive 2009/22/EC to actions brought before that date) and art. 22(2) (the rules on suspension or interruption of limitation periods transposing art. 16 apply only to redress claims based on infringements occurring on or after June 25, 2023); neither paragraph limits how old an infringement may be for the injunctive limb. The distinction between the injunctive and the redress limb is the difference between an exposure that runs against the installed base of terms automatically and one that has to be built out of a national cause of action, and it is the point most summaries of this Directive elide. Art. 21 of Directive (EU) 2024/2853 resolves the cross-reference: references to the repealed Directive 85/374/EEC are to be construed as references to the new one. Annex I point (1) therefore carries Directive (EU) 2024/2853 for products placed on the market after December 9, 2026, and Directive 85/374/EEC for products placed before it. EUR-Lex ↗↩
When a government rescues a bank or rewrites a securities law, it can change a cross-border deal that has already closed. Deal documents usually address that power (see change-in-law, force majeure or bail-in clauses) but cannot fully anticipate sovereign intervention. Virág Blazsek of the University of Leeds, who writes on bank bailouts and central bank digital currencies (CBDCs), discusses how the power works and why the law struggles to keep up with it.
By Eric Martin, CICL Fellow, Center for International and Comparative Law
In conversation with Virág Blazsek, University of Leeds
Updated September 29, 2026, with the interviewee’s written revisions. Revision record.
Governments create and shape markets. Many cross-border deals depend on a government, which may issue the securities, guarantee against losses, or grant the legal permission that makes the transaction possible. The documents set out the parties’ obligations in detail and address sovereign risk through provisions such as change-in-law, force majeure, and bail-in clauses. Those provisions cannot fully anticipate what a government may do to a deal after it closes.
Virág Blazsek practiced law inside financial institutions, as in-house counsel at OTP Bank Plc. in Budapest and at the United Nations Joint Staff Pension Fund in New York, before joining the University of Leeds, where she is an Associate Professor of Law. Her research compares how the United States, Britain, and the European Union rescue failing banks, and she also writes on the digital transformation of financial services. Not long ago she asked a traditional British bank about its savings accounts. “Many times, customer support cannot provide you with accurate information,” she said, “because the people who work for the banks struggle with complex terms and conditions as well as keeping up with the changes.” Trust, on which financial services run, in her view, is “not a legal thing. That’s why it’s so difficult to approach it as a legal scholar. You just know that it’s a necessary ingredient.” She points to David Landes’s The Wealth and Poverty of Nations, Francis Fukuyama’s Trust, and Rachel Botsman’s Who Can You Trust?.
The unwritten sovereign term, coined by the author of this column, refers to the power a government holds over a deal that the documents do not mention, the courts may not be able to review, and the parties did not price. In most cases, regulatory risk looks forward, because a new law affects the next deal. However, this risk looks backward, because the government can use its power against a deal that has already closed. Blazsek does not use the phrase, but her work documents the problem.
Why the Rules Break Down in a Crisis
“Almost every day a 500-page report comes out,” Blazsek said, only half joking. “I consider frequent legal and regulatory changes a pathology of law. When the volume and pace of reforms exceed the capacity of legal subjects, supervisors, and regulators to understand, internalize, and coherently apply the rules, that undermines legal certainty, systemic coherence, and trust.” Since the 2008 global financial crisis, she argues, regulators have issued rules faster than banks and their lawyers can learn them, leaving a body of law built from one-off fixes that do not fit together. The problem is worse in a crisis when a government facing a bank failure acts on its own terms, so no one can say how the rules will operate under stress (see the measures related to the 2023 US FinTech bank failures and the Swiss Government’s management of Credit Suisse’s failure in the same year). Blazsek argues that the post-2008 bank resolution frameworks were intended to replace bailouts with creditor loss allocation (bail-in), but that crisis conditions and political pressures have undermined the credibility of those frameworks over the past decade (see the 2016-17 Northern Italian bank bailouts, Banco Popular’s 2017 resolution in Spain or the above-mentioned 2023 bank failures).
The Rules Behind New Financial Products
On September 15, the World Bank’s International Development Association priced a $3 billion five-year bond, its first dollar global bond sold without registration with the Securities and Exchange Commission. The sale depended on a federal statute that exempts IDA’s securities from registration. The exemption also removes a remedy: with no registration statement, there is no claim under Section 11 of the Securities Act, under which an issuer is liable for a material misstatement in the statement without proof of fault. The bond documents set out what IDA owes; they do not say what happens to the buyer if Congress changes the statute.
Newer products, not yet tested by any crisis, raise the same issue. On September 24, British banks in an industry pilot completed the first live customer transactions using tokenized sterling deposits. UK Finance lists seven participants in the initiative: Barclays, HSBC UK, Lloyds Banking Group, Monzo, Nationwide, NatWest, and Santander. One open question, Blazsek said, is whether deposit insurance should cover these tokens; at this stage, she thinks it should.
Technology deconstructs and reconceptualizes the concept of money. “Two or three years ago, when I started to research CBDCs, I was very much against them,” she said. Her objections were the record of central banks, “which I don’t think is a success story,” and the tendency of governments to keep power once they have it: “The state has the nature of not pulling back when it gets any power in an area.” She worries that a crisis in privately issued digital assets would give central banks a reason “to do what they have always done in crisis situations, to increase their powers.”
She now accepts that tokenized payments can be faster, cheaper, and more flexible, and she thinks digitalization may eventually bring “much more efficiency and clarity,” an outcome she called “optimistic.” What she calls “very alarming” is a central bank taking over the business entirely. Whether a digital currency would help fight inflation is, she said, “a question for myself, as well, currently.”
She would test these systems before legislating for them. Her example is Student Safe (2023-25), a children’s savings app that Hungary’s central bank ran as the first real-world retail CBDC pilot in a European Union member state. The platform combined payments, savings, and financial education tools. Families joined by contract, so no new law was needed: “When it comes to piloting, my opinion is that you don’t need new legislative powers.” The central bank assumed functions normally performed by commercial intermediaries to understand operational realities: “The pilot revealed operational burdens and that the central bank would need to hire more staff to run it every day.”
How Switzerland Rescued Credit Suisse
Deal documents handle regulatory risk through conditions to closing; once a transaction has closed, however, those protections might be exhausted, and a bank failure or sovereign intervention may affect the parties’ rights without a contractual mechanism to reallocate the resulting losses. Blazsek considers the post-2008 resolution frameworks progress, but she says they “only work in business-as-usual times,” because in a panic “there’s no time, political intention, or economic rationale” to follow them.
The bondholders’ complaint, which the Second Circuit accepted for jurisdictional analysis on appeal, alleges that on March 15, 2023, Swiss officials told Credit Suisse: “You will merge with UBS and announce Sunday [March 19, 2023] before Asia opens. This is not optional.” Within four days, an emergency ordinance had waived the shareholder vote, authorized a federal loss guarantee of up to CHF 9 billion, and allowed FINMA, the Swiss regulator, to write roughly CHF 16.5 billion of Credit Suisse’s Additional Tier 1 (AT1) bonds down to zero. In October 2025, the Swiss Federal Administrative Court held that the bonds’ own conditions for a write-down had not been met and that FINMA lacked an adequate legal basis. The court revoked FINMA’s order but has not yet decided whether to reinstate the bonds, and the ruling is on appeal. Blazsek, who expects mass litigation after emergency rescues, was not surprised: “I recall the litigation arising from Banco Popular’s 2017 resolution, which was the first major application of the EU’s post-2008 resolution framework triggering one of the largest waves of litigation in the EU’s history. That ongoing case remains the leading judicial test of the EU’s post-2008 resolution framework.” The EU General Court’s 2022 account of the test cases records approximately 100 actions.
Switzerland had an alternative. The Financial Stability Board later reported that a resolution of Credit Suisse “was ready to be implemented on Sunday 19 March if so decided.” The authorities chose the merger with UBS, which did not prove resolution unworkable. It showed that the framework did not command the confidence of the officials responsible for using it.
The bond contracts did not settle the loss either. Agreeing to a write-down in specified circumstances is not agreeing to one ordered outside them. Blazsek doubts that investors understand such terms. Of the bail-in clauses in European bank debt, she said: “I’m not sure that global investors fully understand the potential consequences of these clauses.” EU law generally requires banks to include contractual recognition of bail-in clauses in many liabilities governed by non-EU law. As a result, an investor purchasing New York-law notes issued by a European bank will often encounter an acknowledgment that the notes may be written down or converted in a resolution proceeding. As she put it, “if you don’t accept it, you won’t make a deal.”
Investors who held $372 million of Credit Suisse’s AT1 bonds through the Depository Trust Company sued Switzerland in New York. A foreign state cannot be sued in an American court unless the Foreign Sovereign Immunities Act provides an exception, and the investors relied on the one for commercial activity: Switzerland, they argued, had brokered the merger as an investment bank would.
In July, the Second Circuit affirmed the dismissal of the case. Switzerland’s loans and guarantees were arguably commercial, the court acknowledged, and at times it negotiated with UBS like a private party. Viewed as a whole, however, its conduct was sovereign. In Judge Calabresi’s words, Switzerland “strongarmed Credit Suisse in a manner that only a sovereign could,” and it changed the law to complete the deal “as no private party could.”
The court did not decide whether the write-down was lawful. It held only that the commercial-activity exception did not give an American court jurisdiction. Legality and remedy are separate questions: the bondholders may win in Bern and still have no court in New York in which to sue Switzerland.
Three Questions for the Buyer
Deal lawyers already price regulatory risk, but mostly the risk that the law will change before closing. The unwritten sovereign term operates after closing. Governments also leave it unwritten on purpose, a practice central bankers call constructive ambiguity: a rescue promised in advance encourages the risk-taking that makes rescues necessary. A buyer cannot negotiate the term into the contract and has to find it instead. Three questions help.
Where does the government sit in the structure? A single government may issue securities, regulate the target, guarantee its obligations, and write emergency law, and its roles can change over the life of the deal. Map each role at signing, at closing, and after.
What powers does the government hold that the documents do not mention? Look for emergency authority, bail-in powers, investment screening, and exemptions with no end date, and ask what would survive of the documents if the government used each one.
If the government acts, where can the buyer seek a remedy? Assume a court later finds the intervention unlawful, and ask whether any court will hear a claim against the government.
Apply the questions to a payments company, a common target in cross-border technology deals. Its business runs through sponsor banks, settlement systems, and accounts that hold customer funds. A government regulates each of them and can take emergency action against any of them, and the purchase agreement may never say what happens to the business if it does. The unwritten sovereign term cannot be negotiated into that agreement, but a buyer who asks the three questions before signing can identify it, and will know what it is actually buying.
Where the documents are silent, the parties fall back on trust, which Blazsek considers essential and hard to study: “We know that it’s true. But what do you do with that?” Tokenized deposits and CBDCs will make the question harder, because ultimately, each depends on government powers that no crisis has yet tested. Blazsek expects the transition to take decades. “I would be really interested to see the tipping point,” she said. “Financial services will look very, very different in 30 years. Technology may eventually help mitigate cyclical financial crises in the distant future, but during the transition it is more likely to increase systemic complexity and exacerbate the next crisis.”
Eric Martin, The Sovereign Rewrite, Conflicts & Capital (Sept. 29, 2026), https://conflictsandcapital.netlify.app/chair/the-sovereign-rewrite.
Conflicts & Capital
The Brief
The lead analysis of an issue: one development in the law of cross-border deals, what it changes for a transaction, and what a deal team should do about it. Newest first.
The trigger looks through the buyer to whoever ultimately controls it. The standard for stopping a deal does not move, so what the widening delivers is a class of deals exposed on the one call-in clock the Regulation gives no end.
Conflicts & Capital
Cornerstones
Longer studies, written to be cited. Each carries an abstract, keywords, and full citations. Newest first.
The clause the buyer negotiated cannot reach the claim that matters. The clause it acquired, sitting in the target’s end-user terms, is one a stranger can bring an action about, and the action reaches versions the target retired before the deal.
Conflicts & Capital
The Practitioner’s Chair
Interviews with the people who shape cross-border deals, on the record and by name. Newest first.
01
The Practitioner’s Chair Law stated as of Sept. 28, 2026
Virág Blazsek of the University of Leeds on how a government that rescues a bank or rewrites a securities law can change a cross-border deal that has already closed, and why deal documents cannot fully anticipate sovereign intervention.
Every interviewee sees and approves their own quotations before a piece runs. The full terms are on the Contact page.
The data hub
By the Numbers
Two datasets on the law that gates cross-border technology deals: what reaches a jurisdiction once a claimant has won, and how long the state there can reopen a closed deal. Release 02 carries an explicit confidence cell on top of that. Where a source gives no figure, the cell records why rather than inferring one.
By the Numbers publishes this column’s own datasets. Each pairs the charts with the full sourced table and a downloadable file at a stable address, licensed for reuse.
Release 02 · How long can a closed deal be reopened?
54 jurisdictions · one question · 42 rows read from a government’s own site, 12 from a reproduction, every one of the 54 named in a source_tier column Law stated as of August 21, 2026Sources retrieved August 21, 2026
A survival period is a number the parties choose. A call-in period is a number the state chooses, and it is the one that decides how long a closed deal stays contingent. This release asks a single question of every jurisdiction in it: after completion, for how long can a screening authority still reach back and unwind or penalize the deal?
4 of 44
Jurisdictions with a power that can reach a completed transaction, and that also give the buyer a date, fixed by the parties’ own transaction, after which the general call-in power lapses. The other 40 do not. The measure counts review, unwinding, divestment, nullity by operation of law, and penalty exposure together, because each of them reaches a closed deal; the table and the file keep them apart.
Release 02, post-closing-reach-2026-08.csv. 54 jurisdictions surveyed; 10 have no post-closing screening power at all, leaving 44 that can reach a completed transaction.
Almost nothing closes. Four regimes give a buyer a date that runs from an event fixed by the parties’ own transaction and expires: Australia, Germany, Hungary, and Slovenia. Ten more publish a period and then extend it, disapply it, or leave an untimed nullity sitting behind it. Twenty-seven have no limitation period at all. Ask instead which regimes leave the deal clean, which is what the phrase invites, and the four are not the answer to that question. Australia drops out of it, because its date closes the review power and leaves the last-resort powers open behind it. Belgium and Cyprus come into it, because their periods lengthen for the deal that should have been notified and then still close, at completion plus five years in both. On that second test the answer is five of 44, and only three of the five advertise a window at all.
Two regimes sit close to the line, and the rule that places them is worth stating. Belgium runs two years from completion and extends to five where there are indications of bad faith. That is a period extended in exactly the case a buyer cares about, so it belongs with the windows that have a hole in them. Hungary runs five years absolutely from the transaction under both its regimes, with shorter limbs of 18 months and 6 months running from the authority’s knowledge; a limb shorter than the absolute bar caps the authority rather than extending it, so it is not a hole and Hungary keeps its date. Cyprus sits on the other side of the same line: its power is bounded at 15 months for an investment outside mandatory notification and at five years where a notifiable investment was not notified, both running from completion, so the gap lengthens the period without removing it. It is the only one of the ten that does. The full rule is in how this was counted.
Post-closing reach, by jurisdiction, of 54 surveyed. n = 54. Shading runs light to dark with exposure; the grey bar is a different kind of answer, not a point on that scale. Each row is classified from the national instrument: the official gazette, the consolidated statute, or the responsible regulator’s own page. Retrieved August 21, 2026.
What this does to the reading of Article 4(4)
Article 4(4) of Regulation (EU) 2026/1386 reads naturally as an extension of national reach: from January 17, 2028, every Member State will have to let its screening authority act on its own initiative for at least 15 months after completion, and may go to five years. It is stated in the future here because this release’s own rule is that a regime adopted but not yet applicable is scored on what governs today. Measured against what Member States actually have today, the floor is the smaller half of the story. 16 of the 27 Member States have no outer limit at all right now, or none on anything but a fine. 15 of those 16 would take the five-year ceiling as the first outer limit their regime carries, and the row-by-row comparison below sets out which, and why Finland is the sixteenth.
But the ceiling does not reach the case this release is about. Both the 15-month floor and the five-year maximum in Article 4(4) are confined to an investment not subject to a prior authorization requirement. The deal that should have been notified and was not is governed by Article 4(5), which requires at least 24 months and sets no maximum whatsoever. So the outer limit the Regulation fixes belongs to the deal that never needed authorization, and it expressly declines to fix one for the deal that needed authorization and did not file. The deal that was filed and cleared is outside both provisions, and what remains open to it is a question for the national regime rather than for Article 4. That is the same asymmetry the ten regimes below have each arrived at on their own, and the Regulation ratifies it rather than closing it.
What Article 4(4) does to the twenty-seven, row by row
Run the floor and the ceiling against each Member State’s own outer limit today, which is what this file records. Article 4(4) governs the investment not subject to a prior authorization requirement, so the comparison is against the outer limit on the general call-in power and not against what happens to a deal that should have been filed and was not. On that comparison the Regulation is, for most of the Union, a contraction rather than an extension.
Where the ceiling bites
Each Member State’s outer limit on the general call-in power, today and from January 17, 2028.
n = 27. Computed from post-closing-reach-2026-08.csv against Reg (EU) 2026/1386, art. 4(4), which requires at least 15 months and permits a maximum of five years for an investment not subject to a prior authorization requirement. The open mark is the outer limit today, the filled mark the limit the Regulation will impose, and a grey mark alone means the two coincide. Finland is drawn at zero because outside the defense and security sector it has no call-in power at all; the untimed power its row records sits on the authorization track, which Article 4(4) does not govern.
Fifteen Member States record no outer limit at all, and the count of regimes Article 4(4) contracts is smaller than fifteen. The fifteen are Austria, Bulgaria, Croatia, Estonia, France, Greece, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Romania, Spain, and Sweden. Two go the other way, and they are the two the floor was not written for: Portugal’s opposition period is thirty days, and Finland has no call-in power outside the defense and security sector, so a 15-month floor creates exposure where none runs today. The remaining ten already sit inside the band: Belgium, Cyprus, Czechia, Denmark, Germany, Hungary, Ireland, Poland, Slovakia, and Slovenia.
Why fifteen is not the number of regimes the ceiling reaches. Article 4(4) times the authority’s power to act on its own initiative over an investment that was not subject to a prior authorization requirement. Most of the untimed exposure in the fifteen is not that. It is what follows from closing without an authorization that was required, which is Article 4(5) territory, where the Regulation fixes no maximum. On the cells as published, only Bulgaria, Romania, and Sweden record an untimed power that is not conditioned on a missing filing, and Lithuania’s reads either way. Three more sit in a nullity that operates by law rather than in a power somebody exercises: Latvia’s regime is purely suspensory with nullity from the moment of the transaction, Spain’s invalidity runs with no window on it, and France makes the commitment void while leaving injunctions and fines untimed, and whether a ceiling on the power to open a review reaches a nullity nobody has to invoke is not obvious. The supported statement is that Article 4(4) puts a first outer limit on the general call-in power in at least three of the 27 and at most fifteen, and that this file does not resolve which. Resolving it needs a fresh reading of each national instrument on the point, which this release did not do.Reg (EU) 2026/1386, arts. 4(4) and 4(5); post-closing-reach-2026-08.csv, columns outer_limit_months, outer_limit_unnotified_months and call_in_post_closing.
Article 4(5) moves nothing at all. Its floor is 24 months for the deal that needed a filing and did not make it. Every one of the 27 already meets or exceeds it: Slovenia sits at exactly 24, Belgium and Cyprus at 60, and the other 24 are at five years or carry no outer limit. So the provision written for the compliant deal changes fifteen Member States, and the provision written for the deal that never asked changes none. The Regulation caps the exposure that was already computable and leaves the exposure that was not exactly where it found it.
The ten regimes whose window has a hole in it
Ten jurisdictions publish a fixed period and then, in the one case a buyer would most want it, take it away, extend it, or leave something untimed standing behind it. Each gives a clean number, and in the case that matters the exception is what governs. A summary that reports the period without the exception reports a number that does not apply to the deal the buyer is worried about.
A date, with a hole in it: the ten regimes that publish a period and then qualify it. Select a column heading to sort.
Jurisdiction
The period as reported
What governs an unnotified deal
United Kingdom
5 years from the trigger event, and 6 months from the Secretary of State becoming aware
Section 2(3) of the 2021 Act disapplies the five-year longstop where a notifiable acquisition completed without approval.
Ireland
15 months from completion for a transaction that was not notifiable
For a notifiable transaction that was not notified, the later of 5 years from completion or 6 months from the day the Minister first becomes aware. Because the limbs are joined by the later of, late discovery extends the reach rather than being capped by it.
Czechia
5 years from completion
No limit at all where the mandatory application was not filed.
Slovakia
2 years from completion
No limit where a critical investment closed without authorization.
Denmark
5 years from completion
The five years was scoped from the outset to the voluntary track. A deal inside the mandatory regime was never within it.
Canada
5 years after implementation
The five years applies only to the residual category in s. 2(c) of the Regulations. A notifiable investment sits in s. 2(a), where the clock ends 45 days after a certified date that never arises if nothing was filed.
Poland
5 years from the acquisition
Article 12e(2) bars only the opening of ex officio proceedings. An acquisition made without the required notification is null and void by operation of law under art. 12k(1), with no period attached.
Portugal
30 days from completion
Article 4(1) runs the 30 days from completion or from the date the transaction becomes public knowledge, should that occur afterwards. An undisclosed deal has no clock running at all.
Belgium
2 years from completion
Extended to five years where there are indications of bad faith. The trigger is a discretionary assessment the buyer cannot compute at signing, which is the defect this release’s classification rule exists to catch. This row rests on the European Commission’s notified list rather than on the Moniteur belge, and says so in its own confidence cell.
Cyprus
15 months from completion for an investment outside mandatory notification
Extended to 5 years from completion where a notifiable investment was not notified. Cyprus is the mildest case here: the gap widens the period but still closes it, so even an unnotified deal reaches a safe date.
Sources, in table order: National Security and Investment Act 2021 (c. 25), s. 2(3); Screening of Third Country Transactions Act 2023 (No. 28 of 2023), s. 12(2)(a) to (c), the two rules in this row sitting in different sub-paragraphs of the same subsection; Act No. 34/2021 Coll., s. 8; Act No. 497/2022 Coll., s. 11; Investeringsscreeningsloven, LOV nr 842 af 10/05/2021, s. 14; Investment Canada Act, RSC 1985, c. 28 (1st Supp.), s. 25.2, with the National Security Review of Investments Regulations, SOR/2009-271, s. 2; Act of 24 July 2015 (Poland), arts. 12e(2) and 12k(1); Decreto-Lei n. 138/2014, art. 4(1); the Belgian Cooperation Agreement of 30 November 2022 on the screening of foreign direct investment, assented to by the Law of 14 February 2023, Moniteur belge, 7 June 2023, taken here from the Commission’s notified list rather than from the gazette and flagged as such in the row’s own confidence cell; N. 194(I)/2025 (Cyprus), s. 3(8)(b) and (c). Every provision sits in the downloadable file with its own row and its own confidence cell.
Set the notified deal against the unnotified one and the gap is the price of not filing. On 34 of the 44 there is no gap at all, because the two columns give the same answer, and on 30 of those 34 the answer is that nothing expires. The ten with a real gap are the ten in the table above, and in every one the buyer who did not file is worse off.
What the filing decision is worth
Every jurisdiction with a power to reach a completed deal, on the one axis the release is about.
Outer limit on the state’s power to reach a completed deal, in months, for a deal that was notified (filled) and one that should have been notified and was not (open). n = 44, the jurisdictions with a post-closing power. Australia’s untimed last-resort power is drawn as a diamond, and is carried by residual_power_untimed rather than by either plotted column. Plotted from outer_limit_months and outer_limit_unnotified_months in post-closing-reach-2026-08.csv. The axis is capped at 120 months; everything past the break is unbounded rather than long.
The four dates that hold
The four regimes where a period runs from an event the parties fix and closes. Select a column heading to sort.
Jurisdiction
Outer limit
What the period does not cover
Australia
10 years
Ten years from the day the action was taken, in reg. 60A of the Foreign Acquisitions and Takeovers Regulation 2015 rather than in s. 66A of the Foreign Acquisitions and Takeovers Act 1975. Read the inversion: the last-resort powers in Part 3 Division 3 of the Act reach a cleared action where a risk has since arisen or false information was given, and carry no express outer limit, so in Australia the date protects the buyer who never filed and not the buyer who notified and was cleared. It is the only regime in this release that runs that way round.
Germany
5 years
Five years from conclusion of the obligating contract, with two months from the ministry’s knowledge, both in s. 14a of the Außenwirtschaftsgesetz. A notifiable transaction is provisionally invalid under s. 15(3) of the same Act but becomes effective if it is not prohibited within those periods, so for a notifiable transaction the five years cures the unnotified deal rather than leaving it open. That cure runs through s. 15(3), not through the period alone.
Hungary
5 years
Five years absolutely from the transaction under both regimes, with shorter limbs of 18 months and 6 months running from the supervising body learning of the deal. The short limbs cap the authority; they do not extend it, which is why this row sits here rather than with the windows that have a hole in them.
Slovenia
2 years
Two years from conclusion of the legal transaction, the takeover bid, or court registration, which is the bar on opening a review and is what the column sorts on. Once a review is opened, the expert group has a further period to deliver its opinion, so the exposure runs past the two years even though the power to start does not. An unnotified deal is not void; the Act imposes fines only.
Outer limit sorts on months, and it is the bar on opening a review, not the point at which a review already opened must end. Germany is the cleanest case in the release, and for a precise reason: for a notifiable transaction s. 15(3) AWG makes the deal effective once the s. 14a periods pass without a prohibition, so there the period does not merely bar a procedure, it perfects the transaction. The other three bar the procedure and leave the transaction where it was.
Sources, in table order: Foreign Acquisitions and Takeovers Regulation 2015 (Cth), reg. 60A, inserted by F2020L01568, with the Foreign Acquisitions and Takeovers Act 1975 (Cth), s. 66A and ss. 79A to 79K; Außenwirtschaftsgesetz of 6 June 2013, ss. 14a and 15(3), with the Außenwirtschaftsverordnung of 2 August 2013, ss. 55 to 62; Act LVII of 2018 (Hungary), ss. 8(2) to (4) and 9, with Government Decree 246/2018, and Act L of 2025, Chapter IV; Zakon o spodbujanju investicij, Uradni list RS 13/18 as amended by ZSInv-C, Uradni list RS 65/23, ch. VI.a, arts. 31.a to 31.e. Every provision sits in the downloadable file with its own row and its own confidence cell.
Six things the buckets do not capture
The United States has no limitation period on the review. Safe harbor attaches only to a transaction on which the Committee or the President has concluded action, and a transaction never notified never enters that category. The three-year rule in the regulations carries a Chairperson carve-out, which makes it an internal escalation rule rather than a bar: no agency notice under § 800.501(c)(1) may be filed more than three years after the completion date “unless the Chairperson of the Committee, in consultation with other members of the Committee, files such an agency notice.” The one-year period in § 800.501(c)(2)(ii) is narrower still, applying only where a foreign person is not an excepted investor by reason of § 800.219(d). The civil penalty is a different question, and it is answered outside the screening chapter: see the bullet below.Section 721 of the Defense Production Act of 1950, 50 U.S.C. § 4565(b)(1)(D) and (E); 31 C.F.R. §§ 800.501(c)(1), (c)(2)(ii) and (d), and 800.219(d).
This release measures the screening instrument only. Every row is classified on what that instrument says. A general limitation statute elsewhere in the same legal system can still bound the penalty, and in at least two of the untimed rows it does. In the United States, 28 U.S.C. § 2462 bars an action for the enforcement of a civil fine or penalty commenced more than five years after the claim accrued, subject to any contrary Act of Congress; whether and how it reaches an administratively imposed CFIUS penalty is a live question this column has not resolved. In Austria, the penalty in section 26 of the Investitionskontrollgesetz is an administrative penal provision, so the general limitation periods in section 31 of the Verwaltungsstrafgesetz 1991 apply to it. Neither displaces the unwind power, which is what this release measures, and neither is recorded in the classification. Comparable general provisions are candidates in Italy, under Law 689/1981, art. 28, in South Korea, under art. 23 of the Framework Act on Administrative Sanctions, and in the Netherlands, under art. 5:45 of the Algemene wet bestuursrecht, and none of the three was closed. A reader pricing the fine rather than the unwind should treat the untimed bucket as unmeasured, not as unbounded.
Switzerland has no foreign-investment screening regime in force. Its Investitionsprüfgesetz was adopted on December 19, 2025 and the referendum period expired on April 17, 2026, but art. 25(2) leaves commencement to the Federal Council. When it starts, the unwind power will be untimed and only the fine will prescribe, at five years. Lex Koller, the separate control regime on the acquisition of real estate by persons abroad, is in force and is outside the scope of this release.Bundesgesetz über die Prüfung ausländischer Investitionen (Investitionsprüfgesetz, IPG) of 19 December 2025, BBl 2026 31, arts. 19 to 21 and 25(2); Bundesgesetz über den Erwerb von Grundstücken durch Personen im Ausland, SR 211.412.41.
Croatia and Cyprus both legislated in late 2025, and the Commission’s list of notified mechanisms does not yet show it. Croatia’s Zakon o provjeri stranih ulaganja, Narodne novine 136/2025, has applied since November 13, 2025 and gives the Ministry of Finance an untimed power to open control proceedings over an unnotified investment, with forced divestment and no fines at all. Cyprus’s Law has applied since April 2, 2026 and is the mildest window-and-gap regime here. Both rows are classified from the national instrument, because the Commission’s list is a record of what Member States have told Brussels and not a statement of what their law says.
18 jurisdictions name artificial intelligence in their screening scope and 18 name semiconductors, and they are not the same 18. Fifteen name both. Greece, Latvia, and Sweden name artificial intelligence and not semiconductors; Japan, the Netherlands, and South Korea name semiconductors and not artificial intelligence, and all three of those have domestic fabrication capacity. The union is 21 of the 46 jurisdictions that have an instrument to name anything in; the other eight have no screening instrument at all, so their cells read not applicable rather than no. Romania is not in either count: its 2026 ordinance is reported to name both, and it could not be read on a gazette, so both cells read not verified. Counted on one rule: named in the instrument or in any sector list, designation notice, or implementing act made under it. Japan does not name artificial intelligence but does name integrated-circuit manufacturing, in the core-sector notice of August 16, 2024 made under the Foreign Exchange and Foreign Trade Act; China names neither in art. 4 of the Measures for the Security Review of Foreign Investment and would reach both only through that article’s open-ended key-technology heading.Foreign Exchange and Foreign Trade Act, Act No. 228 of 1949, arts. 27 and 28, with the core-sector notice of 16 August 2024; Measures for the Security Review of Foreign Investment, NDRC and MOFCOM Order No. 37 of 2020, art. 4.
Two of these figures are not in the instrument a reader would look in first. Australia’s ten years sits in regulation 60A of the Foreign Acquisitions and Takeovers Regulation 2015, not in section 66A of the Act. Canada’s five years sits in section 2 of SOR/2009-271, not in section 25.2 of the Investment Canada Act.
How this was counted
One question, one classification. Every jurisdiction is placed in exactly one of five buckets, read off the instrument rather than off a summary. A date, and it holds: a period runs from an event fixed by the parties’ own transaction and closes. A date, with a hole in it: such a period runs but is disapplied, extended, or outlived by something untimed for a deal that should have been notified and was not. Only the fine is timed: a limitation period found inside the screening regime bounds the power to penalize and none bounds the review power. No limit at all: no limitation period inside the screening regime bounds the review power. Whether one bounds the fine was searched only where a candidate provision surfaced, so all 27 rows in this bucket carry not_searched on penalty_limitation_months. The bucket is a finding about the review power and an open question about the fine. No power to reopen: no post-closing screening power to speak of.
The last two answer the same question, and one rule decides between them. A jurisdiction that times its fine and not its review sits in only the fine is timed; one that times neither sits in no limit at all. Read together they are the 30 jurisdictions in which nothing the state can do to a closed deal expires.
One row reaches a closed deal without a screening power at all, and it is classified on the question rather than on the instrument. India has no post-closing call-in: approval is a condition precedent, so a deal closed without it is not reviewed, it is a contravention of the Non-debt Instruments Rules and so of FEMA, exposing the parties under s. 13(1) to a penalty of up to three times the sum involved and to confiscation. FEMA carries no limitation period for a contravention; its only limitation provision, s. 49, bars cognizance of offences under the repealed Foreign Exchange Regulation Act, 1973 and does not reach the 1999 Act. India therefore sits in no limit at all and not in no power to reopen, because the question asked here is whether the state can still reach the deal after completion, and in India it can, indefinitely, by a route that is not a screening review. It is the only row in the file where the untimed reach is a contravention rather than a review power.
A period whose outer limit runs from knowledge is not a window; a shorter limb that runs from knowledge is not a defect. This is the distinction the release turns on, and it has to be stated in both halves. A clock that starts when the authority learns of a transaction gives a buyer nothing to compute at signing. But a short knowledge-triggered limb sitting inside a long absolute bar can only reduce the exposure, so it does not disqualify the absolute bar. On that test the Netherlands, whose clocks under the Vifo Act all run from the minister’s knowledge, has no window; Bulgaria, whose two years under ch. 6 of the Investment Promotion Act runs backwards from a Commission opinion, has none; Singapore’s two-year lookback under the Significant Investments Review Act 2024, which runs backwards from the target’s own conduct, is not one either; and Hungary, whose outer bar under Act LVII of 2018 is five years from the transaction, does have one. Note also that the qualifying event is the one the parties fix, which is not always completion: Germany’s five years runs from conclusion of the obligating contract and Slovenia’s two from conclusion of the legal transaction, the takeover bid, or registration. All are computable at signing, which is the point.
The law in force, not the law enacted. A regime that has been adopted but has not commenced is scored on what governs today. That is why Switzerland is recorded as having no power to reopen, and why Iceland is recorded as untimed on the eight-week stop in its 1991 Act rather than on the six-month and five-year periods in the Act that applies from January 1, 2027.
Sources. The source_url, source_tier, and confidence cells record each row’s provenance and verification limits. Twelve rows rest on a reproduction rather than on a government’s own site, and the source_tier column names which kind for every row so a reader can recompute the split instead of parsing 54 sentences: nine on an intergovernmental reproduction, being the Commission’s notified list for Belgium, Germany, and Portugal, UNCTAD’s investment policy monitor for Greece, Nigeria, Romania, Serbia, and Taiwan, and FAOLEX for Vietnam; two on a research reproduction, for Israel and South Korea; and one on a private reproduction of the gazette, for Cyprus, whose government PDF could not be opened. Each also says so in its own confidence cell, which is published in full rather than reduced to a rating. The European Commission’s list records what Member States have told Brussels and may lag national legislation. Belgium’s confidence note expressly relies on that list. Germany and Portugal report a rereading of the national instrument, but their source links still point to the list. Direct national-source verification for Belgium and direct national-source links for Germany and Portugal remain to be completed before further reuse.
Rows that rest on an absence. Most of the 27 no limit at all rows rest on having read the relevant chapter and found no limitation provision. That is how these statutes are built, but a negative can be defeated by a provision sitting somewhere the reader did not look, so where a row rests on an absence its note says so. Two such absences were later filled from outside the screening instrument, in both cases against the penalty rather than the review power: 28 U.S.C. § 2462 in the United States and the Verwaltungsstrafgesetz in Austria, neither of which appears anywhere in the screening instrument the row was read from. The classification measures the power to reopen and unwind, so neither row moves, but the exposure is stated here rather than left for a reader to find. Twenty-five rows have not had that second look, and each of their notes says so.
The limits of the classification. It measures the law on the face of the instrument, and three things follow. It measures the general call-in power, so a regime whose general power lapses may keep a residual or last-resort power that does not, which is exactly the inversion Australia shows. It says nothing about enforcement practice: whether an authority in any row has ever exercised its post-closing power, how often, or against what kind of deal is not in the file and was not counted, so a long window in a jurisdiction that has never used one and a long window in a jurisdiction that uses them routinely are the same entry here. And it sorts on time rather than on consequence: whether the exposure at the far end is an unwind, a forced divestment, a nullity by operation of law, or a fine is recorded in each row’s own cells but does not move a jurisdiction between buckets. A reader pricing severity rather than duration should read down the rows, not across the chart.
Six columns are coded rather than written out, so the file can be sorted, plotted, and checked.outer_limit_months gives the outer bound on the general call-in power as an integer, or the sentinel unbounded, or no_power; outer_limit_unnotified_months gives the same for a deal that should have been notified and was not, and it is where the ten holes become computable. Australia’s inversion is not in that column, because both of its cells read the same ten years: the inversion is carried by residual_power_untimed, the only yes in the file. regime_status separates no regime from a regime adopted and not yet commenced and from an advisory body that cannot bind, which the single bucket no power to reopen ran together: on that column Switzerland is adopted_not_commenced, Ukraine and Israel are advisory_only, and seven are none. penalty_limitation_months carries the period on the fine where one was found and the value not_searched where it was not.
gap_kind says how a fixed period fails for the deal that should have been notified, in seven values rather than one: disapplied, never_applied, later_of_knowledge, clock_never_starts, void_by_operation_of_law, extended_on_conduct, and extended_but_closes. An extension is not a disapplication, and the values keep them apart. residual_power_untimed marks a power outside the general call-in that carries no outer limit of its own; Australia is the only yes in the file, and the value is not searched for the other thirteen rows that have a period at all. Every number in these columns is read off the same post_closing_detail and call_in_post_closing cells that were already published;
Where this will move. Japan’s amending Act was promulgated on June 5, 2026 and takes effect on a date to be fixed, no later than June 5, 2027; its outline creates a new power over investments that were not subject to prior notification, so that row will need re-cutting. Switzerland commences in 2027 and Iceland on January 1, 2027. Romania’s 2026 emergency ordinance could be read only in the Senate’s published copy, because the official gazette refused automated retrieval, so its date and gazette citation remain unverified and its classification is left as recorded. India’s 2026 amendment took effect from the date of the FEMA notification, which could not be verified officially.
Every column, and what its values mean
Twenty-four columns, and a reader should not have to infer any of them from the prose. Three sentinels are used throughout and they assert different things. not applicable means the question does not arise for that row, usually because there is no instrument in which it could. not searched means the question arises and was not put. not verified means the answer was found in a source that could not be confirmed against a gazette or a regulator’s own page.
Data dictionary, post-closing-reach-2026-08.csv, 54 rows.
Column
Values
What it records
jurisdiction
54 unique
The row key. Taiwan appears here and not in Release 01, which follows the HCCH and UNCITRAL status tables.
region
7 values
European Union 27, Asia Pacific 11, Europe non-EU 7, Middle East 3, Africa 2, Latin America 2, North America 2. Grouping only; nothing is computed from it except the exclusion of EU rows from the intersection count.
notified_to_commission
yes, not applicable, or a sentence
Whether the mechanism appears on the Commission’s list of notified screening mechanisms read for this release. not applicable is every non-EU row. Two rows carry a sentence because the national instrument postdates the list.
mechanism
free text
The instrument by its own name.
official_citation
free text
The gazette or official citation, or not verified where none could be opened.
mandatory_notification
free text
Whether a filing is compulsory, and above what threshold.
in_force_from
date or a sentence
Commencement. Blank on one row, Iceland, whose Act is adopted and not commenced; the reason is in that row’s note. Where a date could not be pinned to a gazette the cell says so rather than estimating.
ai_named, semiconductors_named
yes 18, no 23, not applicable 8, not verified 2, not specified 1, plus 2 rows of qualified prose
Whether the sector is named in the instrument or in any sector list, designation notice, or implementing act made under it. Hungary and Ireland carry a sentence rather than a code because both are no in the national text and in scope by cross-reference to art. 4(1) of Regulation (EU) 2019/452; the counts treat them as no.
The classification everything else in this release is built on: the five buckets described above, one per row. window is a date, and it holds; window+gap is a date, with a hole in it; sanction is only the fine is timed; untimed is no limit on the review; none is no power to reopen.
post_closing_detail
free text
The one-line statement of the position, in the words the tools on this site render.
call_in_post_closing
free text
The full reading, with the provisions relied on. Its opening word is prose, not a code: three rows open “no” for three different reasons.
provision_refs
free text
The provisions actually read, article by article.
source_url
52 distinct
The address the row was read at. Two pairs of rows share a source.
source_date
2026-08-21 on all 54
The day the sources were retrieved.
note
free text, non-empty on all 54
What the reading leaves open, including anything that contradicts a commonly cited figure.
confidence
free text, 42 distinct
How far the row can be relied on, and why.
outer_limit_months
integer, unbounded 30, no_power 10
The outer bound in months on the general call-in power for a deal that was notified or never needed to be.
outer_limit_unnotified_months
integer, unbounded 38, no_power 10
The same for a deal that should have been notified and was not. Six rows carry a finite value here.
Separates an absent regime from one adopted and not yet commenced and from an advisory body with no binding power.
penalty_limitation_months
not_searched 51, 60 on 2, 24 on 1
A limitation period on the fine, where one surfaced. Fifty-one rows were not searched for it, so this column supports a finding about three jurisdictions and no inference about the other 51.
gap_kind
7 kinds, plus not applicable 44
Why a window does not hold: disapplied 3, clock_never_starts 2, extended_on_conduct 1, extended_but_closes 1, never_applied 1, later_of_knowledge 1, void_by_operation_of_law 1.
residual_power_untimed
yes 1, not searched 13, not applicable 40
A power outside the general call-in that carries no outer limit of its own. Australia is the only yes; the thirteen not searched rows are the other rows that have a period at all.
Every row carries its own call_in_post_closing paragraph with the operative provision, its provision_refs, the official URL it was read from, and a confidence column that says plainly what could and could not be verified.
Eric Martin, By the Numbers, Release 02: How Long Can a Closed Deal Be Reopened?, Conflicts & Capital (Aug. 21, 2026), https://conflictsandcapital.netlify.app/data/post-closing-reach.
Release 01 · The Cross-Border Enforcement Network
116 entries · seven instruments · 812 status cells · read from five official status tables and three further primary sources on one day, recorded against every row Law stated as of August 20, 2026Sources retrieved August 20, 2026
A deal team choosing between a court and a tribunal is choosing between two enforcement networks, and they are not the same size or the same shape. This release reads five multilateral instruments off their primary status tables for one fixed set of 116 entries, and adds the two regional regimes that carry the judgment traffic inside Europe: Brussels Ia inside the European Union, and the 2007 Lugano Convention with Switzerland, Norway, and Iceland. One fixed sample is what lets a jurisdiction be compared across all seven in a single row. It is a defined sample and not a census, and how this was counted says which jurisdictions are in it and which totals are complete.
An award is backed by one obligation of near-universal span. A judgment is backed by several small ones and a great deal of national law. Of the 115 states and territories in the table, 113 are bound by Article III of the 1958 New York Convention to recognize an arbitral award as binding and enforce it, subject to the Article V refusal grounds and to any Article I(3) reciprocity or commercial reservation. Thirty-two have the 2019 Hague Judgments Convention in force. That is 98 percent against 28 percent on the same denominator. Counting the 2005 Choice of Court Convention too, whose Article 8 is itself an enforcement obligation, 40 of the 115 have one of the two, which is 35 percent; adding Lugano makes it 42, or 37 percent. 35 percent is the wide measure and 28 percent the narrow one, and this release quotes both. The gap is wider outside the table. The New York Convention binds 172 states worldwide, against roughly 193 UN Member States. The 2019 Convention binds 32, and every one of them is here: the 33 entries in that column, which count the European Union alongside the 26 Member States its single accession binds, are the whole of it.1958 New York Convention, arts. I(3), III and V; HCCH status table for Convention No. 41.
One asymmetry sits behind both, and it runs the other way. The award’s near-universal span is available only to a party who bought it before the dispute, in writing, over a subject matter both the seat and the destination treat as arbitrable, and only against the people the clause binds. The judgment network asks for none of that. It is there for a claimant with no clause, against a non-signatory, in tort, in fraud, on a guarantee nobody signed, and in an insolvency. In a post-closing dispute that matters more than it sounds: the arbitration clause in the purchase agreement does not bind the target’s customers, the escrow agent, a tax authority, or a co-investor who signed a different document, so the 98 percent covers one leg of the tree and the national law covers the rest. 98 percent is the span of a route the parties must have contracted for; 28 percent is the span of a route that is there whether they did or not.
Read that as a statement about treaty architecture, not about collectibility. A treaty is not the only route to recognition. Where none is in force the question falls to the destination’s own recognition law, which this release does not measure and which has to be checked one destination at a time. Inside the European Union it travels better than an award does: under Brussels Ia a Member State judgment is recognized in the other 26 “without any special procedure being required” and is enforceable there “without any declaration of enforceability being required,” and on a closed and narrower list of grounds: Articles 45 and 46 permit refusal only on application of an interested party, on public policy, defective service of a default judgment, irreconcilability with another judgment, or conflict with the insurance, consumer, employment, or exclusive-jurisdiction rules, and Article 45(3) forbids any review of the jurisdiction of the court of origin outside those heads. Article 23(4)(a) of the 2019 Convention expressly stands aside for rules of that kind adopted before the Convention was concluded.
The asymmetry is not all one way. Article V(1)(e) lets an enforcing court refuse an award that has been set aside at the seat, and Article V(1)(a) lets it reopen the validity of the arbitration agreement itself, so the award’s near-universal span is conditioned on a supervisory court the parties selected once and cannot revisit. What the award has that the judgment lacks is a single obligation a deal team can price at signing, in one line, for almost anywhere. What the judgment has instead is a patchwork that has to be priced destination by destination, which is a harder and more expensive diligence problem, not an impossible one.Regulation (EU) No 1215/2012, arts. 36(1), 39, 45, and 46; 2019 Convention, art. 23(4)(a); 1958 New York Convention, art. V(1)(a) and (e).
Entries where each instrument is in force, of 116 surveyed. The New York, Service and Evidence Conventions are open to states only, so those three bars have a maximum of 115 rather than 116, and the European Union row can appear only in the two Hague judgments bars. Sources: HCCH status tables for Conventions 41, 37, 14, and 20, and the UNCITRAL status table for the 1958 New York Convention. Retrieved August 20, 2026; the Macao SAR rows for the 1965 and 1970 Conventions were read from China’s declarations to the depositary rather than from the status-table header.
The joint distribution, which is the reason for a fixed sample
One row per entry, one column per instrument, sorted by how many of the five bind it.
Number of the five multilateral instruments in force for each entry. n = 116. Grey is not in force, which is not the same as not a party: it carries the six 2019 signatures and the five 2005 signatures that have not been ratified, the United States among them, and the file distinguishes them in its own cells. Computed across the five multilateral columns of cross-border-enforcement-network-2026-08.csv; Brussels Ia and the 2007 Lugano Convention are recorded in the file but not plotted here.
What the table shows
The United States signed the 2019 Judgments Convention in 2022 and has not ratified it. It is party to the New York Convention. For a U.S. counterparty the treaty position is therefore asymmetric, but the practical position is not binary: recognition of a foreign money judgment in the United States is a question of state law, governed by a version of the Uniform Foreign-Country Money Judgments Recognition Act where one is enacted and by common-law comity descending from Hilton v. Guyot where none is. What the arbitration clause buys is uniformity and a known list of refusal grounds. What it does not buy is the difference between collectible and not.Hilton v. Guyot, 159 U.S. 113 (1895); Uniform Foreign-Country Money Judgments Recognition Act (2005), superseding the 1962 Act where enacted. The number of enacting states is not printed here because no count could be verified on the Uniform Law Commission’s own pages.
Six jurisdictions have signed the 2019 Convention without ratifying. Signature is not reach.
Counted the way a deal team asks the question, the asymmetry is far larger than the headline. An entry-wise count answers whether an instrument is in force in a place. A deal team asks whether a judgment from there can be enforced here, and both Conventions work only when both ends are bound. Of the 13,104 ordered pairs among the 115 states and territories in this file, an award has a New York Convention route on 12,650, or 97 percent. A judgment has a treaty route, counting the 2019 Convention, Brussels Ia and the 2007 Lugano Convention, on 1,212, or 9 percent, and on 430 more where the parties made an exclusive choice of court under the 2005 Convention, which is 12 percent at the outside. The entry-wise pair of figures at the top of this release is the conservative statement of its own claim, and the Path Finder below runs the pairwise version one pair at a time.Computed from cross-border-enforcement-network-2026-08.csv over all ordered pairs of the 115 states and territories, excluding the six ordered pairs among China, Hong Kong SAR and Macao SAR, which are territorial units of one State.
Thirty entries are party to the New York Convention and to nothing else in this file. No judgments treaty, no forum-clause treaty, no 1965 service channel, no 1970 evidence channel. In those thirty the arbitration clause is not the better route, it is the only instrument in this file that touches the dispute, and this file counts five. Regional and bilateral instruments are outside it and are not measured, so read the list as thirty places where the five instruments here reduce to one, not as thirty places with nothing else: Bangladesh, Burkina Faso, Chile, Ecuador, Fiji, Ghana, Guatemala, Honduras, Indonesia, Jordan, Kenya, Malaysia, Mauritius, Mongolia, New Zealand, Nigeria, Panama, Papua New Guinea, Peru, Qatar, Rwanda, Saudi Arabia, Suriname, Tanzania, Thailand, the United Arab Emirates, Uganda, Uzbekistan, Zambia, and Zimbabwe. Counting instruments per row, the distribution is bimodal rather than graded: 30 entries have one, 32 have three, 27 have all five, and only 14 sit at two.Computed from cross-border-enforcement-network-2026-08.csv across the five multilateral columns.
Twenty-seven of the 33 entries with the 2019 Convention in force are one accession counted twenty-seven times. The European Union acceded as a Regional Economic Integration Organization, which binds 26 Member States, and the Union occupies a row of its own. Outside that bloc the Convention is in force in six places: Albania, Andorra, Montenegro, Ukraine, the United Kingdom, and Uruguay. That is a larger concentration than the composition effect disclosed for the 22 added rows, and it cuts the other way: the 22 rows widen the gap this release reports, while counting one accession as twenty-seven rows narrows it. On a contracting-party basis the 2019 Convention binds seven, not thirty-three.HCCH status table for Convention No. 41; cross-border-enforcement-network-2026-08.csv, columns hague_judgments_2019 and brussels_ia_1215_2012.
Service and evidence come apart, and four times more often in one direction than the other. Thirteen entries take service under the 1965 Convention and do not take evidence under the 1970 Convention: Austria, Azerbaijan, Belgium, Botswana, Canada, the Dominican Republic, Egypt, Ireland, Japan, Malawi, Pakistan, San Marino, and Tunisia. Three go the other way: Bahrain, Liechtenstein, and South Africa. In Japan and Canada a party can serve through a treaty channel and has none for taking evidence, which is a discovery-planning fact rather than a nine-point gap between two totals.HCCH status tables for Conventions No. 14 and No. 20; computed from the file.
The two Hague instruments overlap but do not nest. The 2005 Choice of Court Convention is in force in 39 entries and the 2019 Judgments Convention in 33, but eight entries have the 2005 Convention and not the 2019 Convention, and two the reverse. Counting either, 41 of the 116 entries, 35 percent, have some Hague treaty route for a judgment, and the 2005 Convention delivers the judgment as well as the clause: its Article 8 is a recognition and enforcement obligation in its own right. As to what it reaches, Article 3(b) presumes that an agreement designating the courts of one Contracting State is exclusive unless the parties expressly provide otherwise. Article 22 would extend the regime to non-exclusive agreements by reciprocal declaration; Switzerland made one on September 18, 2024 and is so far the only Contracting Party to have done so, so no Article 22 regime is yet operative between any two of them. Asymmetric clauses are treated as non-exclusive by the Explanatory Report of Trevor C. Hartley and Masato Dogauchi, which accompanies the Convention but is a reasoned view rather than treaty text, and courts have divided on it.2005 Convention, arts. 3(b), 8 and 22; HCCH status table for Convention No. 37, declarations column; T.C. Hartley and M. Dogauchi, Explanatory Report on the 2005 HCCH Choice of Court Agreements Convention (HCCH, 2013), on art. 3(b). No paragraph is pinned here because the Report was not read in the original for this entry.
Service and evidence run ahead of enforcement, and they are a different kind of thing. 79 entries take service under the 1965 Convention and 69 take evidence requests under the 1970 Convention. It is easier to start a cross-border case than to collect on one. But these two are channels rather than remedies, and membership is not simply an advantage: where service abroad is required, the 1965 Convention is the exclusive route, so being a party constrains the plaintiff as much as it helps. They are plotted apart from the enforcement instruments above for that reason.
The strongest judgments regime in this table is not a treaty at all. For the 27 rows inside the European Union, Brussels Ia does what none of the Hague instruments does: recognition with no procedure and enforcement with no declaration of enforceability, and a Member State judgment therefore crosses an internal border more easily than an award crosses any border. Lugano gives Switzerland, Norway, and Iceland a full regime with the Union and Denmark, but a weaker one: it kept the exequatur that Brussels Ia abolished, so a judgment there still has to be declared enforceable. These are the two columns added to the dataset in this cut, because without them the file could only be read as a story about treaty coverage, and treaty coverage is not the same as reach.Regulation (EU) No 1215/2012, arts. 36(1) and 39; 2007 Lugano Convention, arts. 33(1) and 38(1); Commission, COM(2021) 222 final, on the parties to Lugano.
Scope is the tighter gate. An “in force” cell states that the instrument binds that jurisdiction, not that it covers the claim in hand. The 2019 Convention does not apply to intellectual property at all, nor to privacy, nor to arbitration and related proceedings. The 2005 Convention excludes the validity and infringement of intellectual property rights other than copyright, leaving only contract claims between the parties, and excludes consumer and employment contracts. In a technology deal, where the asset often is the intellectual property, the reach column is a ceiling and the scope articles are what decide the case. 2019 Convention, arts. 2(1)(l), 2(1)(m) and 2(3); 2005 Convention, arts. 1, 2(1) and 2(2)(n)–(o).
How this was counted
What is in the sample. The table has 116 rows: the 94 Members of the Hague Conference on Private International Law, being 93 states and the European Union, together with 22 further states and territories that are not HCCH Members but recur as counterparties, guarantors, or asset locations in cross-border deals. It is a defined sample, not a list of every jurisdiction in the world, and the proportions below are proportions of it.
The 94 Members are a fixed universe, not chosen here. The 22 additions were chosen, on a soft criterion, and every one of them is outside the 2019 Convention while 21 of the 22, all but Kosovo, are inside the New York Convention. That is a composition effect on the very comparison being drawn, and it runs in the direction of this release’s own thesis: on the 94 HCCH Members alone the figures are 92 of 94 against 33 of 94, which is 98 percent against 35 percent. Dropping the 22 rows therefore narrows the gap by seven points. One caution on that denominator. The European Union is one of the 94, it has the 2019 Convention in force, and it cannot be a party to the New York Convention at all, so it sits in one numerator and is excluded from the other by its nature rather than by its choice. On a states-only basis the 94 becomes 93, the 33 becomes 32, and the 92 does not move. It does not close it, and it does not reverse on the other obvious universe: across the 19 states of the Group of Twenty, all 19 are party to the New York Convention and 4 have the 2019 Convention in force. A reader who prefers the unchosen universe should use the Members figures.
Which totals are complete and which are not. For the two Hague instruments this release is about, the sample is the whole membership: the 33 entries with the 2019 Judgments Convention in force and the 39 with the 2005 Choice of Court Convention in force are the complete membership of each, counted the way the HCCH status tables count it. That is not the same as the list of Contracting Parties, and the difference is the point of the bullet above: the Contracting Parties to the 2019 Convention number seven, and the other 26 entries are Member States bound through the Union’s approval; for the 2005 Convention the figures are 13 and 26. The other three columns are partial by design. The New York Convention binds 172 states worldwide, and 111 of them are in this table, which carries Hong Kong SAR and Macao SAR as rows of their own on top of China; the 1965 Service Convention has 84 Contracting Parties, of which 77 are here, plus Hong Kong and Macao, making 79 rows; the 1970 Evidence Convention has 69, of which 67 are here, plus Hong Kong and Macao, making 69 rows. A jurisdiction’s absence from this table says nothing about its treaty position.
The unit of observation is an entry, not a state. The European Union occupies a row of its own because it is the contracting party. It approved the 2005 Convention and acceded to the 2019 Convention in its own name as a Regional Economic Integration Organization, and it is that approval and accession, not a national ratification, that binds the Member States. The count is 26 in each case, but for different reasons, and Denmark is the reason in both: its position under arts. 1 and 2 of Protocol No. 22 on the position of Denmark, under which it takes no part in measures adopted under Title V of Part Three TFEU and none of them binds it, keeps the 2019 Convention from extending to it at all, while under the 2005 Convention it is bound not by the Union’s approval but by its own accession, in force September 1, 2018. Each Member-State row records which of these applies. Counting the Union alongside its Member States is how the HCCH status tables themselves count, which is why the 33 and the 39 here match the figures those tables publish.
What that does to the ratios. The Union is counted once, like any other row. On a states-and-territories-only basis the 2019 Convention figure is 32 rather than 33, and the 2005 Convention figure is 38 rather than 39. The New York Convention figure of 113 is unaffected, because only states may accede to it and the Union is not a party. A percentage taken over all 116 rows would therefore have one row in its denominator that could never have been in its New York numerator, which is why the comparison at the top of this release is stated on the 115 states and territories, 113 against 32, and the 116-entry counts are used only where the instrument itself counts the Union, as the HCCH tables do for the two Hague judgments instruments.
Territorial units are listed separately. Hong Kong SAR and Macao SAR each occupy a row of their own because neither is a contracting party in its own right, and both are counted alongside China, so the New York, Service, and Evidence totals count jurisdictions of application, not contracting states. The routes differ. The Hague Service and Evidence Conventions apply in Hong Kong through the United Kingdom’s pre-handover extensions, continued after July 1, 1997; the New York Convention applies there by China’s extension from that date, and to Macao from July 19, 2005. Twenty-one further territorial units of application appear on these same status tables and are not rows here: the United Kingdom’s extensions of the 1965 and 1970 Conventions to Gibraltar, Jersey, Guernsey, the Isle of Man, Bermuda, the Cayman Islands, the British Virgin Islands, Anguilla, Montserrat, the Falkland Islands, and others, and the Netherlands’ extension of the 1965 Convention to Aruba. On the stated basis they belong in the table, and the sample carries the two that a technology deal is most likely to meet and not the rest. A related limit sits inside an existing row: Denmark’s Article 21 declaration under the 2005 Convention excludes insurance contracts and provides that the Convention shall not apply to the Faroe Islands and Greenland, so its in force cell is true of Denmark proper and overstates the territorial and subject-matter scope. Both points are recorded in the rows themselves.
Macao is on the same footing as Hong Kong. Both Conventions were extended to Macao by Portugal before the handover, the 1965 Convention with effect from April 12, 1999 and the 1970 Convention from December 14, 1999, and China notified the depositary that each applies to the Macao SAR with effect from December 20, 1999 and that it assumed the resulting international rights and obligations, filing declarations of its own for the territory. None of that appears in the status-table header; it is in the declarations pages behind it, which is a trap worth naming, because a header read on its own puts Macao outside two Conventions it is inside. The Service and Evidence figures here are 79 and 69 for that reason.
“In force” is not “signed,” and it is not “in force between any two rows.” A row reads in force only where the status table gives an entry-into-force date on or before August 20, 2026. Signature without ratification reads signed; not ratified and counts as reach of zero; the United States is the consequential instance. Two further limits are not modeled. Declarations and reservations are not: a jurisdiction that is a party subject to a commercial or reciprocity reservation still reads as a party, so the New York Convention column is the outer bound of reach. That limit is not symmetrical, and the asymmetry runs in the direction of this release’s own thesis. Article I(3) reservations are common and live, while no Article 29 notification is recorded under the 2019 Convention, so the New York figure is the one carrying the unstated haircut. Nor is bilateral effect modeled. Under Article 29 of the 2019 Convention the Convention has effect between two Contracting States, however each became one, only if neither has notified the depositary regarding the other. Under Article 28 of the 1965 Convention the same bilateral opt-in is specific to accession, taking effect only as between the acceding state and those contracting states that do not object within the notification window; under Article 39 of the 1970 Convention accession takes effect only as between the acceding state and those contracting states that affirmatively declare their acceptance of it, which is the stricter of the two regimes. An in force cell therefore states that the instrument is in force for that jurisdiction, not that it is in force between that jurisdiction and every other row.
National recognition law is not modeled. A judgment is recognized in most of the world with no treaty behind it, under statutory or common-law rules that differ by destination. Nothing in this file measures that, so a row reading not a party on every judgments column is a statement about treaties and not a statement that a judgment is worthless there. The two regional columns were added precisely because the largest counterexample, Brussels Ia, was sitting inside the sample and outside the data.
Which instruments are counted, and which are not. The criterion is an open multilateral instrument with a published depositary status table, which gives the five here, plus the two regional regimes that carry the judgment traffic inside Europe. Excluded, and not measured anywhere in this release: the ICSID Convention, which reaches investor-state awards on a different enforcement footing from Article III and is the obvious omission in a release about the span of the award obligation; the 1961 European Convention on International Commercial Arbitration; the 1975 Panama Convention; the 1979 Montevideo Convention; the Riyadh and GCC judicial-cooperation agreements; the trans-Tasman regime between Australia and New Zealand; and every bilateral treaty. Any row here may be inside one or more of those. Where this release says a jurisdiction has one instrument and no other, it means one of these five, and the sentence should be read that way.
Sourcing and currency. Each row carries the same eight links in its source_url column: one status table for each of the five multilateral instruments, the HCCH Members list from which the membership column is taken, and the EUR-Lex texts of Brussels Ia and the 2007 Lugano Convention. In source_date is the date it was read: August 20, 2026 for every row in this release. Sourcing is per row and per instrument, not a separate citation per cell; the 812 figure is 116 entries by 7 instruments, which is the size of the matrix rather than a count of distinct sources.
Status tables move, and an entry into force deposited after that date will not appear until the next release. Where a status table gives no date, either because the jurisdiction is not a party or because the table records none, the date column is left empty and the status column carries the reason. No date is inferred.
Every column, and what its values mean
Data dictionary, cross-border-enforcement-network-2026-08.csv, 116 rows.
Column
Values
What it records
jurisdiction
116 unique
The row key. The 116th is the European Union, which is a contracting party and not a state.
hague_judgments_2019
not a party 77, in force 33, signed not ratified 6
The 2019 Judgments Convention. signed not ratified counts as reach of zero and is a third value, not a shade of not a party.
hague_judgments_in_force_from
date, or blank on 83 rows
Entry into force for that row. Blank means the status table gives no date, because the row is not a party or the table records none.
hague_choice_of_court_2005
not a party 72, in force 39, signed not ratified 5
The 2005 Choice of Court Convention, on the same three-value rule.
hague_service_1965, hague_evidence_1970
party, not a party
Two values only: neither status table carries a signed-not-ratified entry for any row in this sample.
ny_convention_1958
party 113, not a party 3
A party reads party whatever reservations it has entered, so this column is the outer bound of reach and not a measure of it.
ny_convention_in_force_from
date
Entry into force for that row.
brussels_ia_1215_2012
not applicable 88, applies 26, applies by parallel agreement 1, the instrument itself 1
Denmark reads applies by parallel agreement, because the Regulation binds it through the 2005 agreement with the Union rather than of its own force. The European Union row reads the instrument itself.
lugano_2007
not a party 85, in force; through the European Union 26, in force; contracting party 4, in force; own signature 1
Three in-force values, because how a row is bound decides what a judgment from where can rely on. The four contracting parties are the Union, Iceland, Norway, and Switzerland; the 26 are the Member States other than Denmark, which is bound in its own right.
hcch_member
yes 94, no 22
Membership of the Hague Conference. The 94 are the fixed universe; the 22 are the added rows, and the effect of adding them is computed above.
source_url
the same eight links on all 116
Five status tables, the HCCH Members list, and the EUR-Lex texts of Brussels Ia and the 2007 Lugano Convention. Sourcing is per instrument, not per cell.
source_date
2026-08-20 on all 116
The day every table was read.
note
free text, non-empty on all 116
What the row’s status hides: territorial extensions, succession, declarations that a status-table header does not show. There is no confidence column in this release; the note carries that work, and Release 02 codes it separately.
Or load it straight into a notebook. The address is versioned in its name, so this snippet keeps working:
import pandas as pd
url = "https://conflictsandcapital.netlify.app/data/cross-border-enforcement-network-2026-08.csv"
df = pd.read_csv(url)
# The European Union is a contracting party, not a place. Exclude it
# for a states-and-territories basis; keep it to match the HCCH count.
states = df[df.jurisdiction != "European Union"]
print(len(states), "states and territories")
print((states.ny_convention_1958 == "party").sum(), "New York Convention")
print((states.hague_judgments_2019 == "in force").sum(), "2019 Judgments Convention")
# Treaty coverage is not reach. Brussels Ia does more for a judgment
# inside the Union than any Hague instrument does anywhere. Test for
# "applies" rather than against "not applicable": the other two values
# are the Union's own row and Denmark's parallel agreement, and
# counting either as a Member State returns 28 instead of 26.
print((df.brussels_ia_1215_2012 == "applies").sum(), "rows inside Brussels Ia")
Eric Martin, By the Numbers, Release 01: The Cross-Border Enforcement Network, Conflicts & Capital (Aug. 20, 2026), https://conflictsandcapital.netlify.app/data/enforcement-network.
The 2019 Judgments Convention, by contracting party
2019 Hague Judgments Convention: Contracting Parties and signatories, August 2026. Select a column heading to sort.
Party or signatory
Status
In force from
States bound
European Union (Denmark not participating)
In force
Sep 1, 2023
26
Ukraine
In force
Sep 1, 2023
1
Uruguay
In force
Oct 1, 2024
1
United Kingdom
In force
Jul 1, 2025
1
Albania
In force
Mar 1, 2026
1
Montenegro
In force
Mar 1, 2026
1
Andorra
In force
Jun 1, 2026
1
Costa Rica
Signed; not ratified
n/a
0
Israel
Signed; not ratified
n/a
0
Kosovo
Signed; not ratified
n/a
0
North Macedonia
Signed; not ratified
n/a
0
Russia
Signed; not ratified
n/a
0
United States
Signed; not ratified
n/a
0
States bound counts the states to which the Convention extends through that party: 26 for the Union, which is its 27 Member States less Denmark, and one for each state that has ratified in its own right. The column sums to 32, which is the number of states the Convention binds and not the number of rows in this dataset that carry it. The six signatories bind none. The full 116-entry dataset, covering the five multilateral instruments and the two regional regimes, is in the downloadable file above. Source: HCCH, Status table for Convention No. 41, hcch.net ↗, retrieved August 20, 2026.
Look up a jurisdiction
Type a jurisdiction and see whether a judgment, an award, a forum clause, service, or an evidence request reaches it. All 116 entries in the dataset.
Source: HCCH status tables for Conventions 41, 37, 14, and 20, and the UNCITRAL status table for the 1958 New York Convention. Retrieved August 20, 2026.
Both releases, one deal
The two releases answer different halves of the same question. Release 01 measures what reaches a jurisdiction once a party has won: an award, a judgment, a forum clause. Release 02 measures how long the state there can reopen a deal already closed. A deal team needs both answers about the same country on the same day, and no single file gives them. The tool below joins the two.
Method
Every row carries the status tables it was read from and the date it was read. Nothing is estimated or rounded for effect, and where a source gives no figure the cell records that rather than a guess. A chart may plot only rows that exist, sourced, in the underlying table, and every chart ships with its table.
Sources are primary: the HCCH and UNCITRAL status tables, the national statute, the official gazette, the regulator’s own page, EUR-Lex, or the court’s own judgment. A law-firm client alert is a pointer to a primary source, never the source itself.
Each release is dated and versioned. Corrections are logged on the Corrections page and never applied silently. Where a jurisdiction is bound through a regional organization rather than its own ratification, the table records that in its note column.
The Archive
The Archive
Everything published so far, newest first. Each piece is tagged by area, and by jurisdiction, deal term, deal stage, and doctrine.
Conflicts & Capital is a column written for CICL Currents, the blog of the Center for International and Comparative Law at Emory University School of Law. It was founded by Eric Martin, a CICL Fellow and its Editor-in-Chief. The views in it are the author’s own and are not those of the Center, of the Law School, or of Emory University. Its beat is the private international law of cross-border deals: what happens when a company is bought, financed, regulated, sanctioned, or sued across a border. It is where CICL Currents publishes commercial private international law, and it takes submissions across the field, starting with the areas below.
What the column covers
Deals. Mergers, acquisitions, and joint ventures: the clearances a deal must win, and how its terms split regulatory and litigation risk.
Emerging companies and venture capital. Founders who incorporate abroad, foreign funds backing startups, and the screening and outbound-investment rules that now reach early-stage money.
Disputes. Arbitration, foreign judgments, and claims against states: whether a win in one country can be collected in another.
Technology and IP. AI, data, and intellectual property that move between legal systems, and rules written in one place that bind companies elsewhere.
Finance and trade. Sanctions, export controls, and tariffs; frozen assets, bank rescues, payments, and digital assets.
Sovereign risk. Governments as regulators, owners, and counterparties, and what happens when one changes the rules after a deal has closed.
Mission
Conflicts & Capital takes a position on contested readings, says where an argument goes next, and compares jurisdictions row by row on a single published rule. Each piece is pin-cited to the instrument. Each dataset asks one question of every jurisdiction in it and publishes the file.
It does not run to a fixed calendar. An issue appears when there is something worth publishing. Between issues the Wire and the docket recompute against the current date, and the datasets carry their version in the filename, so an update is a new file at a new address and the superseded one stays where any citation to it pointed.
Editorial standards
Every legal proposition in a piece is pin-cited in its footnotes to the instrument, the judgment or the official text, and those citations link out. The glossary and the standing pages summarize; they are not a substitute for the citations in the pieces. Every figure is listed in the Ledger with its source and the date it was checked. Every piece carries a law-stated-as-of date, and every dataset carries a retrieval date. Corrections are logged.
Every figure this column prints, with the source relied on and the date it was checked, is in the Ledger; every defined term, with the provision pinned, is in the Glossary. The figures, the glossary, the docket, both datasets, and a record for each piece are also available as one JSON file.
The Wire, the status tape at the head of every page, runs on dated primary-source milestones and recounts the days against the date of your visit, so a countdown on it is never stale. It is a calendar of what is coming, not a complete record of what has happened. It also carries headlines syndicated from named legal sources, attributed on the tape; those are not Conflicts & Capital reporting and are not verified here.
Primary sources are cited in Bluebook form and link out to EUR-Lex, official reporters, or the issuing authority. Corrections are dated, logged, and linked from any piece they touch.
Citation and disclosure
How to cite a version. Each piece has one canonical address, given in its own cite block, and each dataset carries its version in the filename. A revised release is published at a new address and the superseded file stays where any citation to it pointed, so a citation to post-closing-reach-2026-08.csv keeps resolving to the file that was current when it was made. Cite the file rather than the page where the figure is what you are relying on.
Which date governs what. A piece carries a law-stated-as-of date, being the date the author last read the instruments it rests on. A dataset carries a retrieval date, recorded row by row in the file. Standing pages carry a current-as-of date for that page alone. None of them is the date the page was last edited.
Archiving. A reader who needs a fixed copy should archive the page and the dataset file at the time of citation. Every Ledger entry, the glossary, the docket, both datasets in full, and a record for each piece are also published as one JSON file, which carries the data and the metadata but not the text of the pieces.
Conflicts. The author holds no consulting, advisory, or equity relationship with any party named in a piece or with any firm that would gain from a reading taken here, and no piece has been commissioned, directed, or funded by anyone. The column is unpaid. If that changes, the piece it touches will say so above the line.
How this is made. The author uses large language models to assist with research, drafting, data preparation, and code. Responsibility for the analysis, source verification, and published work remains with the author. A source a model supplies is treated as a lead rather than an authority, and no dataset cell is inferred by a model from another cell.
Masthead
Center. The column is written for CICL Currents, the blog of the Center for International and Comparative Law at Emory University School of Law.
Founder and Editor-in-Chief. Eric Martin, CICL Fellow. Biography and disclosures are on the Contact page.
Published by the author. This site is written and maintained by Eric Martin and is not an official publication of the Center, the Law School, or Emory University.
How to cite
Model citation
Eric Martin, What Rome II Does to Your Governing-Law Clause, Conflicts & Capital (Aug. 20, 2026), https://conflictsandcapital.netlify.app/cornerstone/rome-ii-governing-law.
Cornerstones carry an abstract, keywords, and full citations, and are written to be cited. Two other things here are citable, and both have a form of their own.
A dataset
Eric Martin, By the Numbers, Release 02: How Long Can a Closed Deal Be Reopened? (version 2026-08), Conflicts & Capital (Aug. 21, 2026), https://conflictsandcapital.netlify.app/data/post-closing-reach-2026-08.csv.
Cite the versioned filename rather than the /data/post-closing-reach page, which always shows the current release.
A Ledger entry
Conflicts & Capital, The Ledger, entry r2-primary (checked Aug. 31, 2026), https://conflictsandcapital.netlify.app/api/ledger.json.
Corrections
The corrections policy and the full dated log live on the Corrections page. A correction is noted on the piece it touches as well as in the log.
Privacy
The site sets no cookies, runs no analytics and carries no third-party scripts. What it collects, what it stores in your browser, and how to erase it are set out in the privacy statement.
The record
Privacy
What this site collects, what it keeps on your device, and how to erase it.
The only thing a reader gives this site is the email address typed into the subscription form. Beyond that, the host keeps the ordinary server log of each request; nothing else about your reading leaves your browser, and what the site stores on your device you can erase with one button.
Who is responsible
Eric Martin, who writes and maintains this site, is the controller of the personal data described below. He can be reached at Eric.martin@emory.edu. This site is not an official publication of Emory University School of Law or of the Center for International and Comparative Law, and neither is a controller of the subscription addresses, browser storage or server logs described below. Mail sent to that address is a separate matter, dealt with at the end of this section. Brevo, the email service that sends the confirmation message and each issue, and Netlify, Inc., which keeps a record of each subscription form, act as processors. Brevo is established in France; Netlify’s servers are in the United States, so a subscription address is transferred there.
What is collected, and why
Your email address, if you subscribe. It is used to send the column and for nothing else. The legal basis is your consent, given when you submit the form and confirmed when you click the link in the confirmation email; nothing is sent to you until you click it. You may withdraw consent at any time through the unsubscribe link in every issue or by writing to the address above. The address is passed to Brevo, which sends the confirmation email and each issue and stores the address in its contact database.Brevo privacy policy, brevo.com/legal/privacypolicy ↗. A copy of the signup is also recorded in Netlify Forms, and Netlify screens every submission for spam through Akismet, a service of Automattic, Inc. Both are processors acting on the controller’s instructions, and both are in the United States. Under Netlify’s data processing agreement that transfer takes place on the basis of the EU-U.S. Data Privacy Framework, to which Netlify is self-certified, with the European Commission’s standard contractual clauses (Module Two for a controller customer) and the UK IDTA applying as a fallback if the Framework ceases to cover it; a copy can be obtained from Netlify or by writing to the address above.Netlify data processing addendum, netlify.com/pdf/netlify-dpa.pdf ↗; Commission Implementing Decision (EU) 2021/914, Module Two; Commission Implementing Decision (EU) 2023/1795 (EU-U.S. Data Privacy Framework adequacy). The address is not sold, shared, rented, or passed to any advertiser, analytics provider, or list broker. Addresses are kept until you unsubscribe or ask for erasure, and are deleted within 30 days of either, except that Brevo keeps an unsubscribed address on a suppression list so that it is not mailed again.
Server request logs. Netlify records the ordinary information any web server records for a request (IP address, time, page requested, and user agent) for security and abuse prevention, and retains it under its own retention schedule, which the controller does not set and cannot shorten. The legal basis is legitimate interests in keeping the site available and secure. Those logs are not read for analytics and are not combined with subscription addresses. Mail sent to that address is received through Emory University’s mail system and handled under Emory’s own policies; Emory is not a controller of anything else described here.
Nothing else. There is no analytics script, no advertising, no tracking pixel, no social embed, no third-party font, and no A/B testing. The site carries no third-party scripts of any kind. Syndicated headlines on the Wire are fetched by this site’s own server, so your browser never contacts the source.
What this site stores in your browser
The site sets no cookies. The consent rules on storing information on your device are not limited to cookies, though, so here is everything the site writes and why. Two mechanisms are used, both of which stay on your device and neither of which is ever transmitted to this site or anyone else:
Local storage, for things you choose: cc-theme (light or dark), cc-library (the pieces you save), cc-notes-… (the private margin notes you write), cc-history (the last 10 pieces you opened, with the time you opened each, so the front page can offer to pick up where you left off) cc:lastread (the date of your last visit, so the site can tell you what changed since) and cc-pos-… (how far into a long piece you had scrolled, so the piece can offer to resume where you stopped). Two more record settings rather than reading: cc-tape (whether you paused the Wire), cc-shortcuts (whether you turned the keyboard shortcuts off) and cc-stealth (whether Private Mode is on).
The Cache Storage API, through a service worker, which keeps a copy of the pages and the dataset so the site works offline and loads quickly. The cache is versioned and old versions are deleted on each update.
cc-theme, cc-library and cc-notes-… exist only because you asked for them, by choosing a theme, saving a piece or writing a note; the cache exists so the site loads and works offline. cc-history, cc:lastread and cc-pos-… are conveniences you did not ask for by name: if you would rather not have them, erase them below and the site works without them. Nothing here is used to identify or profile you, and nothing leaves your device. Clearing site data in your browser removes all of it, or you can use the button below, which reaches only what is in this browser, and not your subscription address or the host’s server logs.
Private Mode
Erasing is retrospective, and a reader may prefer that nothing be written in the first place. Private Mode, the ◉ button at the top of every page, turns every one of those writes off. While it is on, the site stores no theme preference, no library, no notes, no reading history, no reading position and no visit date; the features that depend on them do not appear, and the rest of the site works as it does now. The service worker is not registered while it is on, so no page or dataset is cached either. Turning it on erases anything already stored, including any cache an earlier visit left, and a band across the top of the page says so for as long as it is on.
One value is still written, cc-stealth, which records the setting itself and nothing about what you read. For a browser that keeps nothing at all, use a private window.
Your rights
If you are in the EU, the EEA or the United Kingdom, you have the right to ask for access to the personal data held about you, its correction or erasure, restriction of or objection to its processing, and portability, and to withdraw consent at any time. Write to the address above and you will get an answer. You also have the right to complain to your national data protection authority.
Changes
This statement is current as of September 29, 2026. If it changes materially, the change will be noted on the Corrections page.
Contact
Contact
Editorial inquiries, submissions, and conversations about cross-border deals.
General
Write to Eric.martin@emory.edu. The address takes editorial inquiries, submission proposals, corrections, and press and academic questions.
Interviews
If you advise on cross-border deals, venture financings, or disputes and would sit for an on-the-record conversation, or know someone who should, write to that address. Interviews run in the Practitioner’s Chair on the terms below.
On the record, and named. Interviews are conducted on the record with practitioners and scholars who agree in writing to be identified and quoted. Nothing is taken on background or on condition of anonymity, and no composite or unnamed “a practitioner told us” interview appears in this column while its present author writes it.
No audio or video is published, and none is made without consent. No recording is made unless the interviewee has agreed to it, and that agreement is recorded at the start of the recording; several jurisdictions in which an interviewee might sit require the consent of every party to a call rather than of one, California, Pennsylvania, and Florida among them (Cal. Penal Code § 632; 18 Pa. C.S. §§ 5703–5704; Fla. Stat. § 934.03), and the desk does not rely on its own consent being enough. This is a statement of the desk’s practice, not legal advice. A recording made for accuracy is used only to prepare the text, is never published, and is destroyed within 24 hours after publication, or earlier if the interviewee requests it. If no piece is published, the recording is destroyed no later than one year after the interview, or earlier on request.
Publication and updates. The author checks quotations against the interview record and publishes the piece without requiring the interviewee’s prior approval. After publication, the author emails the interviewee a link and invites them to flag anything that needs updating. Express restrictions agreed for the conversation are respected. Corrections are recorded under the column’s corrections policy.
Personal, not institutional. Interviewees speak for themselves, not for their firm, institution, or any client. Nothing an interviewee identifies as confidential or privileged, and nothing the desk has reason to believe is, will be published. The interviewee’s own duties to clients and to an employer remain the interviewee’s; this is an undertaking about what will be printed, not a warranty about what may be said.
By the Numbers
Have a deal term you want measured against the cross-border market? Escrow, survival, an indemnity cap, a foreign-investment condition. Suggest it, and it may run in a future release.
Spotted an error? Write to the same address; the policy and the dated log are on the Corrections page. Press and academic inquiries are welcome.
About the author
Eric Martin is a second-year J.D. candidate at Emory Law School and a Staff Editor on the Emory Law Journal. He is the founder and Editor-in-Chief of Conflicts & Capital. Before law school, he was an associate in J.P. Morgan’s M&A Transaction Execution Group in New York. He writes on cross-border technology transactions, venture financing, and the legal implications of AI in M&A, including in the American Bar Association’s Deal Points and in Directors & Boards magazine. He volunteered for the ABA M&A Committee’s Revised Model Stock Purchase Agreement Task Force, where he has co-authored work on agentic liability in transactions, and he co-authored the ABA Antitrust Law Section’s formal comments to the Japan Fair Trade Commission on its 2026 draft merger guidelines, comparing cross-border treatment of resilience, sustainability, and innovation as pro-competitive effects. He works in the Commercial Division of the New York Supreme Court and in the Georgia State-wide Business Court. At Emory, he is a Fellow of the Center for International and Comparative Law and of the AI and the Future of Work Program.
The record
The Ledger
Every figure this column publishes, with the source it rests on and the date it was checked.
This page lists the figures published in the column, the source each rests on, and the date that source was checked. Dataset counts are recomputed from the published CSV files; other figures are entered with their source when the piece is written.
What the dates mean
The date in the last column is the date the figure was checked, not a warranty that the instrument has not moved since. Treaty status changes. Each row names the status table or the official text it was read from, so a reader can check the position as it stands today.
This page holds figures. Legal propositions are pin-cited in the piece itself, in its footnotes.
Open data
Every figure, the glossary, the milestones, both releases and a record for each piece are published as files, licensed CC BY 4.0 for reuse with attribution. The pieces themselves are all rights reserved, quotation with citation excepted.
/api/corpus.jsonEverything in one document: every figure with its source, the glossary, the milestones, both releases in full, and a record for each piece. It carries metadata about the pieces, not their text.
/api/openapi.jsonAn OpenAPI 3.1 description of the endpoints, including what each field means and the two cautions worth reading before computing anything from the dataset.
/api/articles.jsonOne record per piece, with its canonical path, its law-stated-as-of date, its thesis, and the instruments it reads.
/api/wire.jsonThe dated milestones the Wire counts down to, past and future. Two standing items on the tape carry no date and are not in this file, and neither are the syndicated headlines, which are fetched live.
/api/enforcement-network.jsonRelease 01 as JSON, one object per jurisdiction, for anyone who would rather not parse a CSV.
/data/post-closing-reach-2026-08.csvRelease 02. 54 jurisdictions, 24 columns, including two numeric outer-limit columns, a coded source tier, and a confidence cell per row.
Every defined term this column uses, with the instrument named and the provision pinned. The entries summarize; the citations in the pieces govern.
A term in the body of a piece is underlined and carries a short definition on hover. The entry is here, and every term has an address of its own, so a footnote can point at one.
Contribute
Write for Conflicts & Capital
Analysis of cross-border deals, financings, and disputes, written for the lawyers who advise on them and the students preparing to.
Conflicts & Capital is where CICL Currents publishes commercial private international law. It welcomes contributions in any of its areas from students, practitioners, academics, and other writers with relevant expertise. Pieces should develop a clear argument, support it with primary sources, and explain what it means for a deal, a financing, or a dispute.
Email Eric.martin@emory.edu with a working title, a paragraph describing the argument and its practical significance, and the principal sources. Include your name and, optionally, your professional or academic affiliation. A complete draft is also welcome.
Style and format
Length. Eight hundred to fifteen hundred words, excluding footnotes. Propose a longer piece before drafting it.
File. Submit drafts as a Word document, using 12-point type, double spacing, and descriptive section headings.
Citations. Use Bluebook footnotes with pinpoint citations and links to primary authorities. Identify the source of any data.
Writing. State the argument early, define specialized terms, and connect the analysis to a decision a deal team faces.
Editorial review
Submissions are reviewed for accuracy, originality, clarity, and fit with the column. Selected pieces are edited in consultation with the author and published under the author’s byline. Authors review the final text before publication.
Submit original work and disclose any prior publication, simultaneous submission, or relevant professional interest with your proposal. Please use public sources and exclude confidential client or employer information.
The record
Corrections
Material corrections and clarifications are dated and recorded here, with a notice on the affected article.
Policy
A correction records a factual or legal error and its fix. A clarification records a change that sharpens meaning without correcting an error. Both are dated, recorded here, and noted on the affected article. Typographical fixes that do not change meaning are not recorded.
Datasets are versioned in their filenames. A revised release is published as a new file at a new address, and the superseded file remains where any citation to it pointed.
Log
September 29, 2026: The Sovereign Rewrite
The article now incorporates Virág Blazsek’s written revisions. The abstract clarifies that deal documents commonly address sovereign intervention but cannot fully anticipate it. The discussion of regulatory risk, Student Safe, bank resolution, and contractual recognition of bail-in has been clarified; her quotations and supporting references have been updated. UK Finance’s seven named GBTD participants are listed, distinguishing initiative membership from participation in a particular transaction.
