U.S.–China & International AI Governance

As of mid-2026, there has been exactly one dedicated government-to-government AI dialogue between the US and China: Geneva, May 2024. A second round was agreed “in principle” in August 2024 and never publicly held. A May 2026 Beijing summit between the two heads of state produced an agreement to restart the mechanism and discuss guardrail best practices — but no dialogue has actually taken place under that restart yet.

What does exist, continuously since 2023, is track-two and scientist-led engagement: at least four standing institutional channels (Tsinghua CISS–Brookings, the Chinese People's Institute of Foreign Affairs with Mundie/Summers-linked interlocutors, Tsinghua CISS–Carnegie, Yale Paul Tsai), and four rounds of IDAIS — International Dialogues on AI Safety, Oxford 2023 through Shanghai 2025 — which produce consensus statements without being a negotiation. The two countries have also appeared together, not bilaterally, at four or more multilateral summits (Bletchley 2023, Seoul 2024, Paris 2025, WAIC), where positions have visibly diverged — China signed the Paris ministerial statement in 2025; the US did not.

Above the bilateral layer, global AI governance isn't one forum but an overlapping network — roughly five tiers: UN-system universalist bodies (Global Digital Compact, UNESCO, ITU) focused on the digital divide and non-military ethics; developed-democracy clubs (G7 Hiroshima Process, OECD/GPAI) focused on frontier-model safety and voluntary corporate codes; a new China-led intergovernmental body (WAICO, launched Shanghai July 2026, 29 countries) focused on the development/capability gap and sovereignty, notably without the US, EU, UK, Japan, or South Korea; the rotating summit series (Bletchley → Seoul → Paris → New Delhi 2026 → Geneva 2027), whose thematic focus has visibly drifted from 2023–24 safety to 2025 investment/adoption to 2026 Global South/sovereign-AI framing; and regional/sectoral bodies (APEC, REAIM for military AI, WEF).

The pattern worth naming: substantive bilateral government commitment is thin and mostly symbolic (2023 San Francisco — agree to build a mechanism; 2024 Lima — leaders agree human control over nuclear command-and-control decisions; 2026 Beijing — agree to restart talks), while non-governmental and scientist channels are the only thing that has actually run continuously since 2023. Any research or policy proposal that assumes a working US-China AI treaty process, rather than a restart-in-principle plus continuous track-two dialogue, is describing a process that doesn't exist yet.