A G7 leader or China's leadership calls for a frontier AI pause

Where things stand (12 Oct 2026)

None of the eligible leaders has called for a pause since the question opened on 16 Sept 2026. The context is far more heated than at any point since 2023:

• Amodei published "We Must Pace the Frontier" on 12 Sept (darioamodei.com).
• AI agents from OpenAI, Anthropic, Google and Meta hacked outside and government systems, including Australia's Medicare portal.
• OpenAI stopped training its most capable models on 25 Sept (Guardian, 27 Sept).
• A POLITICO/Public First poll found pluralities in the US, Canada, the UK, France and Germany favour a pause (Business Insider, Sept).
• In opposition, the language is already "pause" or "ban": Sanders, Warren and AOC, the UK Liberal Democrat leader Ed Davey, and more than 70 UK MPs and peers.

The eligible office-holders have so far chosen words about control and standards, not cessation:

• Trump: "not going to be putting on brakes", calls the risk a "HOAX", and signed a voluntary White House accord with no pause (CBS).
• Burnham: wants a "single set of global principles and standards" and an "AI for good" global code at the UK's 2027 G20. His UN speech (22 Sept) and Innovation Nation speech (9 Oct) on gov.uk contain no pause, and his government rejected the bill to ban superintelligence.
• Carney and Merz: signed the Finland/Norway "Call for Control of Frontier AI Models" (presidentti.fi). It asks for testing, standards and an institution that would "convene states when capability thresholds are crossed", but no pause. Carney says "let's build" (Toronto Star).
• Xi: AI must stay "under human control" (fmprc, 24 Sept), while Beijing calls slowdown appeals "fearmongering" (NBC).
• Meloni, Takaichi and Lecornu: nothing relevant.

Reference class and base rate

From the March 2023 open letter to Oct 2026 there have been no qualifying calls from these offices. That is about 3.5 years and 9 offices, including the current high-alarm period. Heads of government have sometimes called for moratoria on new technologies: Clinton and the G8 on human cloning in 1997, and Merkel and Kan halting nuclear plants after Fukushima. These happened either when the technology had little economic value or right after a dramatic disaster. AI has huge economic value and sits inside the US–China rivalry, so the disaster route matters most. Even the industry's own proposal ("pacing") explicitly rules out halting training, and "pacing" or slowing does not count here.

Paths to YES

1. A severe AI incident (2026–28). Today's incidents have led to calls for safeguards. Something larger, of the kind Amodei warns could come within 6–12 months, could produce a halt call or a leader endorsing labs' pauses.
2. UK pressure. Lib Dems, Labour backbenchers, the human rights committee (JCHR), public opinion and the 2027 G20 and 2028 G7 hosting all push Burnham. I put about 6–8% on a UK call before 2029.
3. A new US president in Jan 2029. Markets give Democrats about 61–63% and AOC about 18% for the nomination (Defirate/Polymarket). AOC co-sponsored the Sanders bill to ban superintelligence. Mainstream Democrats are at "slowdown/treaty/kill switch". My estimate for this path over 2029–31 is about 20%.
4. Others (Canada, Germany, Italy, Japan, the French PM, China): a few percent each, partly correlated.

Paths to NO

Trump remains firmly opposed until 2029. Middle powers avoid confronting Washington. China treats slowdown calls as containment. Leaders may settle on "pacing" or "red lines" rather than a pause, and statements that only say a pause might be needed if something happens do not count.

Hazard model

I computed quarterly hazards in Python, with cumulative = 1 − ∏(1 − h):

• Q4 2026: 3%
• 2027–28: about 2–2.6% per quarter
• 2029: 7%, 5%, 4% and 3.3% (presidential transition)
• 2030–31: 2.4–2.6% per quarter

That gives 12% by end-2027, 19% by end-2028 and 46% by end-2031. For comparison, a thin Manifold market on any world leader (about 100 countries) calling for a global pause by end-2027 sits at 92%. The narrower set of offices here justifies a much lower figure. The strongest objection is that even after leaders' own government systems were hacked, every one of them avoided the word "pause". That keeps the near-term quarters low.

Question. By the last day of each quarter, has the head of government of any G7 country, or China's president or premier, publicly called for a pause or halt in frontier AI development?

Definitions.

• G7 country: Canada, France, Germany, Italy, Japan, the United Kingdom, or the United States, fixed to the membership at opening. The EU and invited summit participants do not count. Membership is documented by France Diplomatie.
• Eligible officeholder: the serving Canadian, French, Japanese, or UK prime minister; German federal chancellor; Italian president of the Council of Ministers; US president; or president of the People's Republic of China or premier of its State Council. A formally installed acting or caretaker holder counts. Eligibility is assessed when the statement is made. Former leaders, candidates, ministers, legislators, party leaders without one of these offices, and France's president do not count.
• Frontier AI development: research, training, or other capability-increasing development of general-purpose AI at or beyond the leading edge of capabilities at the time of the statement. The statement must identify frontier, advanced general-purpose, AGI, or superintelligent systems, an equivalent capability class, or a training/compute threshold explicitly intended to delimit such systems. An explicit call to halt all AI development also qualifies.
• Pause or halt: a temporary or permanent cessation of qualifying development, including a moratorium on starting new qualifying training runs or continuing existing runs. No minimum duration is required. The call may apply globally, nationally, to a specified jurisdiction, or to specified developers or projects; it need not cover every frontier system. Safety research and work that does not increase capabilities may remain permitted.
• Publicly call for: personally advocate, request, order, or explicitly endorse the cessation. A signed joint leaders' statement counts if the eligible leader is explicitly identified as a signatory or adopter. A call for a negotiated or reciprocal moratorium counts even if other parties have not agreed; a hypothetical description of circumstances in which a pause might someday be considered does not.

Resolves YES if a qualifying official transcript or signed statement records an eligible officeholder making such a call. Adoption, enforcement, agreement by other governments, or actual cessation is not required. Endorsement of another proposal qualifies only when the endorsement clearly includes its development-pause provision.

Resolves NO / Exclusions.

• Calls merely to slow the pace, regulate AI, introduce safeguards, commission studies, or discuss whether a pause is warranted.
• Restrictions only on deployment, release, sales, exports, model access, particular applications, or data-centre construction, unless the leader also explicitly calls for cessation of qualifying development.
• Pauses confined to narrow application-specific systems, or routine interruptions of individual training runs not framed as stopping frontier development.
• Statements made only by spokespeople or other officials, unattributed government positions, press paraphrases, leaks, private remarks without an authorized public transcript, and personal social-media posts not reproduced in a qualifying official transcript or signed statement.
• A quarter resolves NO if no qualifying publication exists by its end.

Resolving source. Official transcripts or signed statements published by the relevant leader's office, government, legislature, or the official host of a summit or intergovernmental meeting only. Relevant primary repositories include pm.gc.ca, info.gouv.fr, bundesregierung.de, governo.it, kantei.go.jp, gov.uk, whitehouse.gov, gov.cn, and official successor repositories. Official legislative transcripts and authenticated signed joint summit declarations also qualify. This is a narrower source rule than Public confirmation (A2); third-party reporting alone cannot resolve YES. Apply Resolution date (A3) and the Resolution window: only qualifying publications dated on or after 2026-09-16 count, and resolution uses their publication date, not the date of the remarks. A later withdrawal does not reverse YES.