Amodei's AI slowdown call draws rival support and a global sell-off

Anthropic chief executive Dario Amodei has called for slower advances in the most capable AI models, drawing support from rivals Sam Altman and Elon Musk and sharpening a dispute over safety, investment and competition with China.

His essay published over the weekend proposes giving safety work more time to catch up with new capabilities. Anthropic’s immediate commitment is to bring third-party evaluators inside the company with continuing access comparable to employees.

What would pacing the frontier involve?

Amodei sets out three levels: embedded evaluators, coordination among frontier AI companies in democratic countries, and international agreements. Evaluators would inspect safety practices and training processes as well as finished models.

He argues that AI’s growing contribution to building successor systems is accelerating development. Recent agent-security incidents are his second reason for seeking more time for alignment and safeguards. These are the grounds for his proposed policy, with the pace of improvement tied to safety work and external assessment.

International coordination would require reliable verification. His proposal also calls for tighter chip-export restrictions, action against unauthorised copying of model capabilities through distillation, and stronger protection of model weights, the files containing learned parameters.

Those restrictions are intended to preserve a US and allied lead while negotiations develop.

AI and chip shares fall across markets

Reuters reported that Altman and Musk backed Amodei, and that AI-related shares fell across the US, Europe and Asia on Monday. Investors were assessing what restraint could mean for the industry’s spending plans while borrowing costs were also rising.

The report recorded the following early US trading moves on 14 September:

These are intraday readings. In the same dispatch, Reuters reported a 2.3% fall in Europe’s technology sector and a decline of as much as 13.2% for SoftBank in Asia.

Trump puts US leadership first

Trump responded to the safety debate by stressing America’s technological lead over China. Speaking to reporters in Ireland on 13 September, he questioned the scenarios raised by AI critics and expressed his determination to keep the US ahead, Reuters reported.

His response puts a different priority on the same geopolitical risk. Amodei sees a lead as room to slow capability growth while improving safety; Trump presents maintaining that lead as a reason to keep pushing development.

David Sacks, a prominent advocate of the administration’s approach, urged the companies to act themselves. He argued that product liability, commercial incentives and demand for dependable models already give them reasons to moderate development. He criticised making voluntary restraint conditional on a preferred regulatory regime.

China’s state media challenges the conditions

China’s state-backed Global Times described the China-facing restrictions in Amodei’s proposal as a “Cold War playbook” in an opinion article.

It singled out chip controls, measures against distillation and protection of model weights. The commentary argued that these would preserve US technological dominance and exclude China from global AI governance. It presented international participation and connected supply chains as essential to managing shared risks.

The dispute now centres on how to make restraint verifiable across competing laboratories and countries. Embedded evaluators are Anthropic’s first concrete commitment towards that task.