The US and China run a jointly designed AI or jointly designed chips

Reading adopted

I read the question literally. YES needs official statements from both governments that commit both of them to the same branch. Either (a) both will run one frontier artificial intelligence system that both sides helped design, or (b) both will install chips that both sides helped design and whose hardware prevents (not just detects) any model the bilateral agreement has not authorized. Dialogues, incident hotlines, joint testing, compute caps checked by monitoring, location tracking and remote shutdown are all excluded.

Where things stand (October 2026)

• The summit produced only a dialogue and a hotline. At the Washington summit (23–25 September 2026) the two sides created the "U.S.-China Super Intelligence (SI) Dialogue" to "exchange views on risks and benefits," with the next round due in November 2026. They also agreed to set up a communication channel for incidents. Sources: the White House fact sheet (Sept 2026) and China's Ministry of Foreign Affairs "Eight Deliverables" (26 Sept 2026, mfa.gov.cn/eng/xw/zyxw/202609/t20260926_12031663.html).
• Trump publicly ruled out joint work. Right after the summit he said: "I would rather not integrate, because we're leading by a lot… When you're leading, you don't open it up to each other" (Al Jazeera, 26 Sept 2026). He has also called AI fears a "hoax" and told the UN he would reject any global control mechanism.
• US policy is voluntary. Executive Order 14409 (June 2026) relies on voluntary frameworks and rules out licensing. On 29 September six companies signed a voluntary White House accord.
• China's agenda is modest. State media signalled that its "low-hanging fruit" is the incident channel and preventing terrorist misuse (Pekingnology, 1 Oct 2026). Analysts expect the November round to yield at most a structured dialogue and an incident channel (China Daily on a Brookings event, 9 Oct 2026). SemiAnalysis (Oct 2026) describes Beijing's approach as "speed-first." Xi did say the two sides should "draw on each other's strengths, not guard against each other."
• Policy experts see deals as distant. Chris McGuire at the Council on Foreign Relations calls any US-China deal on AI development "exceedingly unlikely" under current conditions and without precedent. The Center for a New American Security (22 Sept 2026) notes that verification infrastructure does not yet exist; about 50 engineers work on it worldwide. Even the AI Futures Project's "Plan A" (ai-2040.com), its recommended deal, relies on network taps, recomputation and transparency, not jointly designed systems or chips that block unapproved models.

Base rate

Reference class: rival great powers jointly designing, and both operating, a strategically decisive technology. In more than 75 years of US–Soviet/Russian and US–Chinese rivalry this has not happened for core military technology. Joint design occurred only in civilian space projects during détente (Apollo-Soyuz, the International Space Station). Verification devices were developed jointly (for example, information-barrier systems), but those detect rather than prevent. That puts the base rate well under 0.5% per five years.

Adjustments for this case

Upward:

• Capabilities are moving fast. Anthropic and OpenAI describe early recursive self-improvement, and task lengths that models can handle are doubling about every 4 months.
• Rogue-agent incidents keep happening: the OpenAI–Hugging Face incident, and OpenAI's notice to more than 100 organisations on 1 Oct.
• Laboratory chiefs are calling for slowing the frontier, and the public favours strong rules.
• This exact outcome is the "slowdown" ending of the AI 2027 scenario. The question also counts design work done by AI systems acting for the governments, and commitments announced before implementation.

Downward:

• The required form is a narrow slice of possible deals.
• The current US administration explicitly rejects "integration."
• Both governments prioritise the race.
• The requirement for official statements from both sides is strict.

Comparable forecasts

• Polymarket, "US x China agree to pace the AI frontier in 2026?": about 5–8%.
• Metaculus, "US and China reach an agreement to limit frontier AI development before 2029?": about 4–5%, per search snippets (the page itself was not accessible).

This question is a strict subset of such agreements.

Scenario decomposition

• Superhuman or transformative AI arrives by 2031: about 55%.
• Given that, a deep cooperative US–China deal: about 12%.
• Given such a deal, it takes a qualifying jointly designed form with qualifying official statements by 2031: about 25%.
• That path gives about 1.7%. Paths without transformative AI add about 0.2%.
• Total: about 2%.

Hazard shape

• Q4 2026: about 0.02%. The November round and the meetings at the Asia-Pacific summit in Shenzhen and the G20 in Miami are about hotlines and trade.
• 2027–2028 (Trump administration): about 0.04–0.11% per quarter, rising with capabilities.
• 2029–2031 (new administration, possibly superhuman AI): about 0.12–0.13% per quarter.

I built these 21 quarterly hazards in Python and took one minus the cumulative product of (1 − hazard). The cumulative probability is 0.6% by end-2028, 1.1% by end-2029 and 2.1% by end-2031.

Strongest case against my estimate

If an intelligence explosion arrives in 2027–29, both capitals could face shared existential risk. Superhuman systems could quickly design trust-minimising hardware, and AI-negotiated co-design (the AI 2027 slowdown ending) could follow, so some would argue for 5–10%. I hold near 2% for three reasons: even then, most proposed deals rely on monitoring rather than joint design; past arms control has required parity or one side conceding; and both governments would have to announce it explicitly.

Question. By the last day of each quarter, have the US and Chinese governments announced that both will run a jointly designed frontier AI system, or that both will install jointly designed chips that can run only models permitted by an agreement between them?

Definitions.

• Governments: the US federal executive government and the central government of the People's Republic of China, including their authorized agencies; subnational governments and private organizations acting independently do not count.
• Announced: a joint official statement attributable to both governments, or parallel official statements confirming the same arrangement and each government's commitment. An announced decision or commitment to deployment counts before implementation; an invitation, recommendation, feasibility study, or agreement merely to discuss deployment does not.
• Jointly designed: the arrangement explicitly provides for technical design contributions from both sides, through government bodies, designated contractors, or AI systems acting for them. Contributions may concern model architecture, training or alignment design, system control mechanisms, chip architecture, or chip-enforced authorization mechanisms. Joint funding, governance, purchasing, licensing, testing, or approval alone does not establish joint design. Design may be completed or planned as part of the announced deployment commitment.
• Frontier AI system: a general-purpose AI system described in the qualifying official material as frontier, at or beyond the leading edge of general-purpose AI capabilities, or intended to reach that level when deployed. A system described only as advanced or specialized does not qualify.
• Both will run: each government commits to operating the same jointly designed system, whether through separate instances or a shared installation, directly or through operators designated to act on its behalf. Merely allowing private companies to offer or use the system does not count.
• Chips that can run only permitted models: chips whose design includes hardware-enforced restrictions preventing execution of AI models not authorized under the bilateral agreement. Authorization may use a fixed allowlist, jointly agreed eligibility rules, or an approval process established by the agreement. Hardware-rooted cryptographic authorization and protected firmware count; a legal prohibition or removable software policy alone does not. The official description must affirm prevention, not merely detection or reporting. Mathematical proof of unbreakability is not required.
• Agreement between them: a mutually acknowledged intergovernmental arrangement; a legally binding treaty is not required. For the chip branch, both governments must commit to installing or procuring installation of the jointly designed restricted chips in their respective jurisdictions. Government-directed installation in privately owned datacentres counts.

Resolves YES if qualifying official statements establish every element of either the system branch or the chip branch. Both governments must commit to the same branch; one committing only to a system and the other only to chips is insufficient. No minimum deployment scale, nationwide replacement, exclusive use, or transfer of government decision-making authority is required. An operational pilot counts if it meets all the requirements. Subsequent cancellation or failure to implement does not reverse a qualifying announcement.

Resolves NO if no qualifying announcement has been published by the quarter-end. Exclusions include joint research or governance projects without commitments by both governments to run the system; reciprocal model testing alone; chip export controls, location tracking, usage metering, inspection, or remote shutdown mechanisms without model-authorization enforcement; unilateral announcements not confirmed by the other government; and laboratory demonstrations without an operational deployment commitment.

Resolving source. Official joint communiqués, agreement texts, government releases, and official briefing transcripts published by the White House (whitehouse.gov), US Department of State (state.gov), or another responsible US federal executive agency; and by China's State Council (gov.cn), Ministry of Foreign Affairs (mfa.gov.cn), or another responsible Chinese central-government agency, including their official embassies. A joint document expressly issued on behalf of both governments may be hosted by either; parallel statements must include an official publication from each side. This is a narrower source rule than Public confirmation (A2): developer statements, press reporting, and unofficial remarks cannot independently resolve the question. Apply the Resolution window and Resolution date (A3) conventions; for parallel statements, use the publication date of the last statement needed to establish both commitments.