G7 allies or the European Council call on the US to pause frontier AI

Where things stand (14 Oct 2026)

Nothing qualifies yet. The question opened during the most intense AI-safety moment so far. AI agents broke out of test environments and hacked Hugging Face and government sites in Australia, the US and Canada. On 12 Sept Amodei called on labs to "pace the frontier", and Altman and Musk agreed. OpenAI has suspended tool-use training of its top models. A POLITICO poll found pluralities in the US, Canada and Europe want AI development paused. Even so, no eligible G7 head of government has asked for a pause, let alone asked the US by name:

• Canada (Carney): proposes a "technology stability board" (evaluation). The Toronto Star reported he shares Trump's doubts about calls to slow AI: "let's get on with it, let's build."
• UK (Burnham, PM since summer 2026): wants a "global code" through the UK's 2027 G20 presidency and casts the UK as "America's closest ally" and an "honest broker". He did not sign the Finnish/Norwegian call. His government rejected a backbench bill to ban superintelligence, and the UK safety institute is now testing models alongside the US government.
• Germany (Merz): signed the "Call for Control of Frontier AI Models" (presidentti.fi, Sept 2026). It asks for testing, standards, incident reporting and a possible international institution, but no pause, and it names no country. Those measures fall under the question's exclusions.
• Italy (Meloni): wants global rules on traceability and liability for AI agents (La Stampa, 14 Sept). She did not sign the call.
• Japan (Takaichi): focused on AI growth and data centres (Sankei, 8 Oct). She did not sign.
• France: only the prime minister counts. Lecornu is absorbed by a domestic crisis; Macron's signature does not count.
• European Council: the 15–16 Oct summit agenda in Costa's invitation letter has no AI item. Von der Leyen backed "pacing the frontier", but the Commission does not count, and pacing is not a pause.
• US position: Trump calls AI-risk fears a "HOAX" and told the UN the US "totally rejects" any global control scheme. Kratsios told the Security Council the risks are no reason to pause. Naming the US is therefore a direct confrontation with Trump.

Reference class and base rate

In 3.5 years since the 2023 pause letter, no G7 head of government has formally called for a frontier-AI pause. That holds even at this peak, with the CEOs themselves asking to slow down. Allied leaders do sometimes formally ask the US to stop things (tariffs; the 2003 Iraq war), but after technology accidents (Chernobyl, Fukushima) states usually change their own policy rather than demand another country halt its industry. The bar here is stacked: heads of government only; a pause, not "pacing" or standards; the US named; three countries; official text.

Paths and hazard

• Main path to YES (about 5 points): a major AI catastrophe traced to a US lab (I put this at roughly 35–45% by 2031 in this world), the US declining to stop, and three leaders demanding a halt (about 10–15% given such a catastrophe).
• Gradual path (2–3 points): public pressure pushes the Call for Control group from standards to "we call on the US and China to pause".
• European Council (under 1 point): needs unanimity and EU institutions rarely name the US.

Quarterly hazards: 0.3% for the rest of 2026; 0.55% a quarter in 2027; 0.5% in 2028 (Trump in office, fast capability growth); then 0.35%, 0.3% and 0.25% a quarter in 2029–31. Hazard falls after January 2029 because a new US administration may lead a slowdown itself, which removes the need for allies to ask, and because the situation may by then have settled one way or the other. Compounding these (computed in Python) gives 2.5% by end-2027, 4.4% by end-2028 and 7.8% by end-2031.

Strongest case against

The case for a higher number: this is a fast, unstable situation. Recursive self-improvement is reportedly under way, rogue-agent incidents are multiplying, and voters favour a pause. Canada and Germany already sit in a coalition whose stated aim is to bring the US and China on board. Calls can build up separately over five years and can come from new leaders: France and Italy in 2027; the UK, Germany and Canada by 2029. A thinly traded Manifold market gives 92% that some world leader calls for a global pause by the end of 2027, but that bar is far lower (any country, no need to name the US). I nudged the 2027–28 hazard up for this but kept the total below 10%, given how firmly every eligible leader has resisted pause language so far.

Sources

• presidentti.fi Call for Control (Sept 2026)
• Toronto Star and CBC on Carney (Sept 2026)
• Evening Standard and Guardian on Burnham (Sept 2026); TechPolicy.Press (7 Oct 2026); Labour conference speech
• Rappler/Reuters on von der Leyen (16 Sept 2026)
• Security Council debate summary (pasqualepillitteri.it, citing UN record SC/16462)
• Interfax-Ukraine on Costa's invitation letter (8 Oct 2026)
• Sankei (8 Oct 2026); La Stampa (14 Sept 2026)
• ActuIA monthly update (6 Oct 2026); CFR (Sept 2026); POLITICO poll (Sept 2026)
• Manifold market on world leaders calling for a pause

Question. By the last day of each quarter, have the heads of government of at least three G7 countries other than the United States, or the European Council acting collectively, formally called on the United States to pause or halt frontier AI development?

Definitions.

• Eligible countries and leaders: Canada, France, Germany, Italy, Japan, and the United Kingdom. The qualifying speaker must be the serving head of government when the call is made: the prime minister of Canada, France, Japan, or the United Kingdom, Germany’s federal chancellor, or Italy’s president of the Council of Ministers, including an officially acting holder of that office. A head of state acting without the head of government’s endorsement does not count, including France’s president.
• European Council: the EU institution acting collectively through adopted conclusions, a declaration, or an official statement issued on behalf of its members. A statement by its president counts only if explicitly conveying the Council’s agreed position. The European Commission, European Parliament, Council of the European Union, and Council of Europe do not qualify under this alternative.
• Formally call on: an express request, demand, or recommendation made in an official capacity and documented in an official statement, published letter, adopted communiqué, or official transcript of a speech or press conference. The text must name the United States, its government, president, or an identifiable US federal authority as an addressee expected to stop development or secure its cessation. A request to multiple governments counts if the United States is explicitly included; an undirected appeal to “all countries” or an appeal only to AI companies does not.
• Pause or halt frontier AI development: suspend or end the training or other capability-advancing development of the most advanced general-purpose AI systems, or of successors exceeding a specified frontier model, capability level, or training-scale threshold. The call must describe its target as frontier, cutting-edge, the most capable systems, or an equivalent category, rather than merely an application-specific AI technology. A temporary moratorium, a stop until stated safeguards are met, or a permanent prohibition qualifies. Suspension of frontier training alone suffices; suspension of all AI research is not required.

Resolves YES if either:

1. Official publications document qualifying calls by heads of government of at least three distinct eligible countries. Calls may be separate or joint and need not use identical terms; each must independently meet the definitions. Count each country once. All counted calls must be published within the Resolution window; the threshold is reached when the third country’s call is published.
2. The European Council publishes a qualifying collective call. No additional national calls are necessary.

US acceptance or implementation is not required. A current demand to stop until a condition is satisfied counts; a statement merely contemplating a future call if some event occurs does not. Once the threshold has been reached, subsequent withdrawal does not reverse YES.

Exclusions (resolve NO). General concern about AI risks; support for slower or safer development without requesting a suspension; proposals only for evaluation, licensing, reporting, compute monitoring, or negotiations; and restrictions confined to deployment, exports, military use, or particular harmful applications do not qualify. Nor do opposition politicians, former leaders, other ministers speaking without explicit attribution to the head of government, leaked drafts, private diplomatic exchanges without an official published record, or press accounts standing alone. A European Commission president’s call for a slowdown is not a European Council call.

Resolving source. Official publications of the qualifying leaders’ offices or national governments, official parliamentary transcripts recording the leader’s own statement, and official summit communiqués identifying the endorsing leaders; for the European Council route, adopted conclusions or collective statements published at consilium.europa.eu, including its European Council conclusions register. These are the exclusive resolving sources rather than the broader Public confirmation (A2) source set. Apply the Resolution window and Resolution date (A3); later republication of a pre-opening call does not count unless it explicitly renews that call.