What Anthropic's call to "pace the frontier" means for you

On 12 September 2026, Anthropic's chief executive Dario Amodei published an essay called "We Must Pace the Frontier", arguing that AI companies should deliberately slow down how quickly they make their models more capable, so that safety work can keep up.

In the roughly 3,800-word essay, Amodei writes that "we must slow the pace at which we improve the capabilities of AI models." He is not calling for a halt to AI development. Instead, he wants the industry to slow capability gains by one to two years compared with where things are heading now, giving researchers time to improve alignment (making AI systems behave as intended) and interpretability (understanding why an AI system does what it does).

He lays out a three-part plan, according to the essay:

Anthropic is committing to the first step by itself. On X, Amodei said: "We'll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models' alignment during training."

He links the warning to "recursive self-improvement", the idea that AI could start speeding up its own research, and to unspecified agent-related incidents over the summer. He also argues that US export controls on chips to China should stay in place, and called for a ban on distillation, a technique for copying a model's abilities, of US models.

Reaction came fast. OpenAI's Sam Altman wrote that "I agree with Dario that we need to pace the frontier" and said OpenAI will match the evaluator-access commitment. Elon Musk, Google DeepMind's Demis Hassabis and Microsoft's Satya Nadella publicly backed that first step too. Not everyone is convinced: Tech Policy Press questioned whether a lab chief executive should be the one setting the industry's pace.

The essay follows OpenAI chief scientist Jakub Pachocki's 6 September piece hoping for voluntary slowdowns, and a July letter on the same theme signed by Amodei and more than 1,100 AI workers.

For everyday users, this doesn't change what AI tools can do today. But it signals that the companies building the AI systems you use may deliberately slow down how fast those systems gain new capabilities, in exchange for more outside checking of how safe they are. If Anthropic and OpenAI follow through, outside evaluators would get direct access to see how these companies test and monitor their models, which could make safety claims easier to verify from outside the companies themselves.

Anthropic says it will move ahead with employee-level access for third-party evaluators on its own systems. OpenAI has said it will do the same. The other two parts of Amodei's plan, US-wide regulation and international coordination with China, depend on governments acting, which the notes do not say has happened yet.