U.S. and China Say They Will Resume AI-Safety Talks After Two-Year Pause

The United States and China have said they will resume official AI-safety talks after a two-year halt. The reported opening, potentially timed around a Trump-Xi summit this month, revives a channel for discussing frontier-model risks—but not evidence of a deal on how to slow the AI race or enforce limits.

China reportedly sees the discussions as an important summit deliverable and wants any frontier-AI limits to apply equally to American and Chinese models. That position goes directly to the hard part of bilateral safety diplomacy: both governments are being asked to coordinate on risks while competing for technological advantage.

A channel shaped by mistrust

Beijing has already signaled concern that safety negotiations could become another mechanism for restricting China’s technological rise. Yuyuan Tantian, a social-media account affiliated with state broadcaster CCTV, warned of that possibility. In July, Xi Jinping said AI should remain under human control while also opposing an overstretched national-security concept and one country putting its security above others'.

Why a sweeping pact is a poor fit

Dario Amodei, Anthropic’s chief executive, has argued that AI progress should slow long enough for safety guardrails to catch up, and that coordination with China is part of that task. But he has also set a demanding condition for any agreement: it must be ironclad in its verification or narrow enough that a violation would not be militarily existential.

The narrow areas advocates see as more plausible

• Bans on using AI for biological weapons and other specific dangerous applications, which Amodei considers potentially feasible.
• Greater mutual visibility into what each government fears about advanced AI and how it manages those risks.
• Informal exchanges alongside official talks: Chinese and Western groups have held about a dozen Track 1.5 and Track 2 discussions since 2023.

An opening, not a shared speed limit

Sam Altman and Elon Musk have expressed support for U.S.-China coordination on AI safety; Altman said a one-page agreement between Trump and Xi should be achievable. Yet a short statement of principle would leave unanswered the operational questions: which systems are covered, what conduct is prohibited, and how either government would establish compliance.

The talks therefore matter less as a sign that Washington and Beijing have resolved their rivalry than as a chance to keep AI-risk communication alive inside it. A limited arrangement on clearly defined dangerous uses would be easier to assess than a broad pledge to manage frontier AI—and would avoid turning safety promises into bargaining chips in disputes over chips, tariffs or technology controls.