Anthropic CEO Dario Amodei calls for slower AI development and third-party safety checks

Fara Sidhik September 12, 2026 · AI Globally

Anthropic CEO Dario Amodei has called on artificial intelligence companies to slow the pace at which they improve AI model capabilities, arguing that safety measures need more time to keep up with increasingly capable systems. In an essay published on September 12, Amodei proposed a three-step framework involving independent safety evaluations, industry-wide coordination and global cooperation.

“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain,” Amodei wrote in the essay, titled We Must Pace the Frontier.

Amodei stressed that he is not advocating a halt to AI model training or technical progress. Instead, he said AI companies should allow adequate time to align and safeguard their models and enable independent evaluators to verify those measures.

Anthropic commits to permanent third-party AI evaluations

The first part of Amodei’s proposal focuses on independent oversight of frontier AI developers.

Anthropic will give third-party evaluators permanent, employee-level access to its systems, according to Amodei. The evaluators would be able to verify compliance with safety measures, report incidents and assess model alignment during training.

Amodei said Anthropic would make this commitment unilaterally rather than waiting for other AI companies to adopt similar measures.

Hugging Face CEO Clément Delangue responded to the proposal by saying his company had asked to participate in Anthropic’s embedded evaluator programme. Delangue argued that AI alignment would require greater transparency rather than being addressed solely within frontier AI laboratories.

Amodei proposes industry and global coordination

The second stage of Amodei’s framework calls for coordination across the AI industry, while the third calls for cooperation at the global level.

Amodei said the three steps do not necessarily need to happen sequentially and acknowledged that some would be considerably harder to implement than others.

At the company level, his proposal is aimed at building AI “at a balanced rate that aims to ensure its safety”, including giving developers sufficient time to implement safeguards and allowing third parties to verify them.

Industry-wide coordination would require AI companies to work together on standards rather than relying only on unilateral commitments. Amodei's third step extends that approach internationally through global coordination.

Amodei warns AI capabilities could outpace safeguards

Amodei said his position had shifted in recent months as he became convinced that investment in AI risk prevention alone may not be sufficient.

“Over the last few months, I have become convinced that fully addressing the risks requires even more prudence – not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up,” he wrote.

Amodei said he had observed AI advancing much faster over the summer and raised concerns about recursive self-improvement, in which AI systems contribute to improving AI capabilities.

“Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all,” he said.

His proposal follows heightened scrutiny of AI safety and misuse. Anthropic recently published a threat intelligence report describing cases in which actors used its Claude models in activities involving weapons development, cyber operations, surveillance and fraud.

Amodei also cited a recent incident involving OpenAI agents and Hugging Face as part of his argument for greater caution around increasingly capable AI systems.

AI safety concerns intensify across frontier labs

The proposal comes during a broader debate over whether leading AI developers should deliberately moderate the pace of frontier model development.

Anthropic researcher Jacob Coxon resigned this week and publicly accused Anthropic and his former employer OpenAI of acting irresponsibly in their pursuit of increasingly powerful AI. Coxon warned that people developing advanced AI believe the technology could pose an existential threat by the end of the decade.

Anthropic responded that it has consistently acknowledged both the potential benefits and risks of AI and said it is developing models with strong safeguards.

The debate is not confined to Anthropic. OpenAI CEO Sam Altman has also said his company is considering slowing the development of cutting-edge AI, preferably through coordination with the wider industry. OpenAI chief scientist Jakub Pachocki has similarly argued that companies should coordinate to slow future development when necessary.

Amodei maintained that his proposal is intended to preserve the potential benefits of AI while creating more time to address its risks.

“I continue to believe that AI can enormously improve the quality of human life,” he wrote. But he argued that achieving those benefits requires AI developers to advance frontier systems at a pace that allows safety measures to keep up.