Foreign Affairs Makes Case for US-China AI Safety Pact
Key insights
- The US and China share AI risk overlap in three areas: bioweapons uplift, non-state actor misuse, and autonomous weapons escalation.
- Authors propose the Incidents at Sea Agreement and IAEA safeguards as concrete structural templates for bilateral AI governance.
- The Trump-Xi summit is reportedly deprioritizing AI governance, pushing authors to call for a separate parallel diplomatic track.
Why this matters
AI practitioners and founders building dual-use systems need to understand that a US-China AI safety framework, if it materializes, would likely impose capability disclosure or verification requirements that reshape export control compliance obligations. The Cold War arms control analogy signals that formal verification mechanisms, not voluntary norms, are the policy direction serious analysts are pushing toward, which has direct implications for model transparency and audit infrastructure. The sidelining of AI governance at the Trump-Xi summit means the window for industry to shape any eventual framework through track-two channels is open right now, before positions harden.
Summary
A Foreign Affairs essay argues the US and China share enough overlapping AI safety interests to sustain formal bilateral cooperation despite deep geopolitical rivalry, drawing explicit parallels to Cold War nuclear arms control.
The core overlap covers three risk areas: bioweapons uplift from AI systems, misuse by non-state actors, and autonomous weapons escalation. The authors propose the Incidents at Sea Agreement and IAEA safeguards as structural templates for a new governance framework.
Essentially: (US government, China) face shared catastrophic-risk scenarios where unilateral action is insufficient.
- The Trump-Xi Beijing summit is reportedly sidelining AI governance in favor of Iran and trade talks, making a parallel diplomatic track more urgent.
- The Cold War analogy is deliberate: arms control succeeded even when broader US-Soviet relations were hostile.
- An IAEA-style model would require both sides to accept some form of AI capability disclosure.
Whether Washington and Beijing can insulate AI safety talks from the rest of the bilateral relationship is the question the essay raises but cannot resolve.
Potential risks and opportunities
Risks
- If AI governance stays off the Trump-Xi agenda through year-end, no institutional escalation channel exists for autonomous weapons incidents before they become military flashpoints
- US AI labs (Anthropic, OpenAI) face compliance uncertainty if a bilateral framework eventually imposes capability disclosure requirements that conflict with current Commerce Department export control rules
- China's Cyberspace Administration could use a formal bilateral framework as legitimacy cover to export restrictive AI governance norms to third-party countries, narrowing the global regulatory baseline
Opportunities
- Track-two diplomacy organizations (RAND, Carnegie Endowment, CNAS) are positioned to host the parallel diplomatic track if official summit channels remain blocked, giving them outsized agenda-setting influence
- AI safety researchers with active US-China academic ties gain leverage as credentialed intermediaries if government-to-government channels stall on verification design
- International standards bodies (ISO TC 42, IEEE SA) could move into the governance vacuum if the bilateral track fails, creating early-mover advantage for companies that engage these bodies before norms solidify
What we don't know yet
- Whether an IAEA-style verification mechanism would require disclosure of model weights, training data, or eval results, something neither government has publicly agreed to
- Which US agencies would lead a parallel diplomatic track on AI governance if it remains excluded from the Trump-Xi bilateral agenda through the rest of 2026
- Whether the non-state actor framing in the essay covers open-source model proliferation, which the US and China currently regulate under incompatible domestic regimes
Originally reported by foreignaffairs.com
Read the original article →Original headline: Foreign Affairs: America and China Can Make AI Safer — Cooperation Is Necessary and Possible