AI systems negotiate or monitor a US–China agreement

Where things stand (as of 13 Oct 2026)

Nothing published since the 16 Sep 2026 opening date qualifies.

• AI talks so far. The US–China AI track is at a very early stage. Treasury Secretary Bessent and Vice Premier He Lifeng agreed on 20 Sep to formalise an AI dialogue and proposed an incident notification line. The Trump–Xi state visit (23–25 Sep) then produced two separate lists of outcomes. Neither was a signed instrument, and neither says anything about AI negotiating or monitoring:
  • The White House fact sheet set up a "US–China Super Intelligence (SI) Dialogue", with the next exchange by November 2026, plus a channel for "SI incidents".
  • China's foreign ministry listed eight "deliverables and understandings". One says the two militaries will conclude a crisis-communication memorandum of understanding (MOU).
• No joint documents yet. A Fudan Development Institute review (8 Oct) counts "zero joint documents" on AI between the two since 2023. It expects any written arrangement through March 2027 to cover only "tier-one" issues, such as what triggers an incident report and who the contacts are, not evaluations or capability limits.
• What's coming. The November dialogue round, the APEC summit in Shenzhen (November) and the G20 in Miami (December) give chances to sign things like an incident-channel MOU. None is expected to give AI any role.
• Political positions. Trump says "when you're leading, you don't open it up". China rejects agendas built on risk warnings. CFR's Chris McGuire calls an AI arms-control deal "extremely unlikely" for now.
• Terminology. Executive Order 14434 (29 Sep) requires US agencies to say "Super Intelligence"/"SI" instead of "AI". The question requires systems "explicitly identified as artificial intelligence". I assume "SI" would count, because the order defines it by the statutory definition of AI. Still, this adds a little ambiguity for US-only statements.

Reference class and base rate

As far as I know, no international agreement has ever been officially credited to AI negotiators. No treaty has ever given AI the job of making compliance findings without human review. Signed US–China instruments of any kind are rare.

Related Metaculus questions, seen only through search snippets because the site blocked direct access:

• Any US–China AI treaty that controls or monitors AI development before 2030: about 12–15%.
• A treaty limiting artificial general intelligence before 2030: about 5%.

This question needs much more than either of those: an explicit official statement about AI's role.

Paths to YES

1. Monitoring branch (most likely path, mostly 2029 onward). A substantive US–China agreement to slow or limit AI, with verification, that uses "AI inspectors". IFP's August 2026 report describes these as a trusted AI that inspects the other side's sensitive data, reports only "not breached", then wipes its memory. A design like that is close to an explicit statement that no human reviews each finding. The AI Futures Project's verification plan also mentions AI-automated monitoring.
   • My rough estimate: about a 10–15% chance of such a deal by 2031, times about a 15–20% chance that its text explicitly makes AI findings autonomous. That gives roughly 2%.
2. Negotiation branch. An official statement that AI agents exchanged substantive proposals, for example on technical tariff schedules, or that Trump said so in a published transcript. This is plausible mainly if AI capabilities take off very fast: roughly 1–1.5%.
3. Other routes. Long-shot multilateral or trade instruments: about 0.5%.

Paths to NO, and the strongest case against my number

• Both governments' AI statements stress human control (the 2024 Lima nuclear statement; China's "maintain human control" line in its September readout).
• Each side distrusts AI built by the other, and AI inspectors can be jailbroken, which IFP acknowledges.
• Verification technology is immature: the AI Futures Project reports that only 4 of 17 needed workstreams have any active effort.
• The required explicit wording is a strong filter on its own.

The strongest argument for a higher number is how fast AI is moving in this world. Anthropic and OpenAI report early recursive self-improvement (AI increasingly building its successors). Anthropic reports that the length of tasks AI can complete is doubling about every 4 months. Lab CEOs are calling for slowing the frontier. By 2029–2031, AI-led negotiation and verification could be normal. That is why my per-quarter chance rises over time and I don't go below about 2%. The strongest argument for lower is the complete lack of precedent plus the strict text requirement, which is why I don't go above about 5%.

Forecast construction

I set a per-quarter chance of the first qualifying publication and accumulated it (computed in Python):

• 0.05% for the rest of 2026;
• 0.07–0.10% per quarter in 2027;
• 0.12–0.17% in 2028;
• 0.20–0.26% in 2029 (new US administration, possible deal window);
• 0.27–0.34% in 2030–31.

Cumulative: about 0.4% by end-2027, 1.0% by end-2028, 1.9% by end-2029, 3.0% by end-2030 and 4.2% by end-2031.

Sources

• White House fact sheet, Sep 2026: https://www.whitehouse.gov/fact-sheets/2026/09/fact-sheet-president-donald-j-trump-advances-a-fair-and-reciprocal-relationship-with-china-while-hosting-historic-state-visit/
• China's foreign ministry, eight deliverables, 26 Sep 2026: https://www.fmprc.gov.cn/eng/xw/zyxw/202609/t20260926_12031663.html
• Executive Order 14434, 29 Sep 2026: https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/
• Fudan FDDI analysis, 8 Oct 2026: https://fddi.fudan.edu.cn/1b/20/c21253a793376/page.htm
• Pekingnology, 1 Oct 2026: https://www.pekingnology.com/p/china-signals-what-it-wants-from
• CFR (McGuire): https://www.cfr.org/articles/how-to-reach-an-ai-arms-control-deal-with-china
• CNAS, 22 Sep 2026: https://www.cnas.org/publications/cnas-insights/cnas-insights-u-s-china-ai-agreements-need-verification-not-trust
• IFP, 6 Aug 2026: https://ifp.org/preparing-for-ai-research-automation/
• AI Futures Project verification plan: https://ai-2040.com/supplements/verification-plan
• Dario Amodei, "We Must Pace the Frontier", Sep 2026: https://darioamodei.com/post/we-must-pace-the-frontier
• Anthropic, "When AI builds itself": https://www.anthropic.com/institute/recursive-self-improvement
• Metaculus question 21219, via search snippet

Question. By the last day of each quarter, have the US and Chinese governments signed an agreement whose text or accompanying official statements say that AI systems conducted the negotiation, or are assigned to monitor compliance without a person reviewing each judgement?

Definitions.

• Governments: the United States federal government and the central government of the People’s Republic of China, acting through authorized representatives. Agreements signed by ministries or agencies on behalf of these governments count; agreements solely between companies, universities, local governments, or political parties do not.
• Signed agreement: a written instrument signed by representatives of both governments, including a treaty, executive agreement, memorandum of understanding, or signed joint declaration. It may be bilateral or multilateral, legally binding or nonbinding, and concern any subject, not necessarily AI. Ratification or entry into force is not required. An official statement explicitly confirming that both governments signed is sufficient evidence of signature even if the instrument itself is not public.
• AI systems: one or more computational systems explicitly identified as artificial intelligence in the qualifying source; no particular developer or capability level is required.
• Conducted the negotiation: AI systems acted as negotiators for at least one government, or as a mediator, making and responding to substantive proposals about the agreement’s terms in exchanges between the negotiating sides that produced the signed agreement. Human direction, oversight, approval of concessions, or final approval and signature do not disqualify this branch. Translation, transcription, research, summarization, drafting assistance, or private simulations and advice to human negotiators alone do not count.
• Assigned to monitor compliance without a person reviewing each judgement: the agreement establishes an AI role in assessing compliance with at least one obligation of that agreement, and the text or official statement explicitly says that the role does not require individual human review of every AI compliance judgement before that judgement is issued or used as an official monitoring finding. Explicit descriptions of autonomous or unreviewed compliance determinations qualify; silence about review does not. Sampling, audits, review of selected cases, appeals, and a human override are compatible with this condition. Automatic enforcement or sanctions are not required.
• Accompanying official statements: official government publications explaining the particular agreement, its negotiation, or its implementation, including subsequent explanations that explicitly identify it. A general statement about using AI in diplomacy or monitoring is insufficient.

Resolves YES if a qualifying source establishes that both governments have signed the same agreement and establishes either the negotiation branch or the compliance-monitoring branch above. The two branches are independent; either alone suffices. A definite assignment of the monitoring role counts before operational deployment, but a commitment merely to study, test, or consider such a role does not. An explicit official attribution that AI systems conducted the substantive negotiation suffices without publication of negotiating transcripts.

Resolves NO if no qualifying confirmation is published by the quarter-end. Exclusions include unsigned announcements or draft agreements; AI recommendations used only by human negotiators; ordinary automated data collection, anomaly detection, or verification tools that do not make compliance judgements; monitoring in which each judgement requires human review before official use; and statements describing only a proposal, exercise, or hypothetical arrangement.

Resolving source. The agreement text as officially published by either government or an official treaty depositary; or official statements, readouts, fact sheets, or published briefing transcripts from the White House (whitehouse.gov), US Department of State (state.gov), the Chinese Ministry of Foreign Affairs (mfa.gov.cn), China’s State Council (gov.cn), or the US or Chinese ministry or agency that signed the agreement. An explicit statement from either government can establish the AI role; matching statements from both are not required. News reporting, vendor claims, personal posts, and unofficial interviews do not qualify. This is a narrower source rule than Public confirmation (A2). Apply Resolution date (A3) and the Resolution window convention; if evidence of signature and of the AI role appears separately, use the publication date of the last necessary qualifying item.