SAN FRANCISCO, USA – Three prominent American AI leaders have endorsed slowing the development of increasingly powerful models, turning an industry safety dispute into a test of international cooperation. Anthropic’s Dario Amodei, OpenAI’s Sam Altman and Elon Musk have expressed support for restraint. Their agreement, however, does not establish that their companies have implemented a coordinated pause.

The appeal follows the resignation of Jacob Coxon, a researcher who worked at OpenAI and Anthropic. He told the BBC that colleagues were “genuinely frightened” by rapid advances and the possibility of losing control. Anthropic defended its safeguards, testing and safety research. Other industry figures questioned whether catastrophic predictions were overstated or served commercial interests. The extinction probabilities discussed in the report are individual judgments, not established scientific forecasts.

Amodei’s proposal combines independent evaluators inside laboratories, common standards among companies in democratic countries and eventual international agreements. He explicitly distinguishes this approach from halting model training. His warning that more capable autonomous systems could cause devastating cyberattacks within six to 12 months is a projected risk, rather than a demonstrated timetable.

Political agreement remains elusive even in Washington. President Trump has rejected the slowdown argument, emphasizing American leadership over China. That division complicates any expectation that Beijing would accept restrictions American policymakers themselves resist.

China’s First Visible Objection Is About Power

A September 13 commentary from China’s state-run Global Times offers a substantive response. It acknowledges that rapid AI development creates safety, ethical and loss-of-control risks, but attacks Amodei’s proposal as an effort to constrain China and preserve American technological dominance. This is a media editorial, not a formal government decision or a commitment from Chinese developers.

The criticism addresses an explicit feature of Amodei’s argument: alongside cooperation, he advocates stronger restrictions on China’s access to advanced chips and measures against unauthorized copying of frontier capabilities. He says international pacing should protect the lead of the United States and its allies. The diplomatic difficulty is clear: a safety arrangement advertised as preserving one side’s advantage will be harder for the other side to trust.

In the public material reviewed through September 13, no direct endorsement or rejection of this specific appeal could be verified from DeepSeek, Alibaba, Baidu, Tencent or ByteDance. Their absence from the verified response record should not be interpreted as a decision to disregard the warnings.

Indeed, China’s safety efforts predate this controversy. Concordia AI’s July 2026 review identifies growing research into autonomous agents and loss-of-control risks, with Alibaba, Tencent, Huawei and ByteDance among the major corporate contributors to safety research. Its findings challenge the assumption that Chinese developers are uniformly indifferent to such dangers.

But the same review identifies a substantial accountability gap: only five of ten leading foundation-model developers published safety evaluation results alongside any release during its reporting period, and none did so consistently. Research activity and public promises therefore cannot substitute for independently assessable evidence.

Competition and Cooperation Can Advance Together

The most defensible expectation is that Chinese companies will continue pursuing commercial and technological gains while selectively supporting safety cooperation. That is an analytical assessment, not an announced collective position.

It fits Beijing’s existing policy: China’s Global AI Governance Action Plan promotes innovation, infrastructure and open-source development alongside risk assessment, testing and international safety collaboration. Continuing development is therefore compatible with China’s stated approach to managing risk.

A genuine American slowdown could give Chinese rivals more time to narrow capability gaps or win customers. The size of that opportunity would depend on what actually slows, for how long, and whether Chinese firms face comparable requirements. Public expressions of concern alone create no measurable competitive windfall.

Elsewhere in Asia, reporting has moved faster than verified corporate commitments. Japan’s FNN covered the appeal on September 13, but that coverage does not establish Japanese industry support. South Korea’s Naver and Samsung, and China’s Zhipu, already joined the Seoul frontier-safety commitments; those earlier pledges should not be presented as responses to this weekend’s announcement.

The practical opening lies in shared testing methods, incident reporting and research that helps competing laboratories identify dangerous behavior. The 2026 Singapore Consensus identifies such cooperation as serving mutual interests. Whether governments and companies will accept meaningful verification remains the decisive question.