Anthropic CEO Dario Amodei is calling on the world’s leading artificial-intelligence companies to deliberately slow the pace at which they improve the capabilities of frontier AI models.
Amodei argues that the industry needs additional time to understand and manage the risks created by increasingly powerful systems.
In a new essay, he writes: “We must slow the pace at which we improve the capabilities of AI models.”
Reuters reports that Amodei has proposed a three-stage framework intended to pace frontier development while stronger safety measures are established.
The significance goes beyond another warning about artificial-intelligence safety.
Anthropic is itself one of the companies competing at the technological frontier.
Its Claude models compete with systems developed by OpenAI, Google, xAI, Meta and other laboratories, while Chinese companies are pursuing increasingly capable models of their own.
That creates a fundamental coordination problem.
Anthropic can conclude that slower capability development would reduce risk.
But if Anthropic slows while its competitors continue advancing at full speed, restraint could simply transfer technological and commercial advantage to companies that choose not to participate.
The same calculation applies to every competitor.
Each company may individually recognise reasons for caution while simultaneously fearing the consequences of being the only organisation that exercises it.
The problem becomes still harder internationally.
Any agreement among American frontier laboratories would have limited effect if major competitors elsewhere continued accelerating development.
That means voluntary restraint may be difficult to sustain even when the companies involved genuinely believe that slowing down would be safer.
The question therefore extends beyond whether AI development is moving too quickly.
It is whether a competitive global industry is capable of collectively reducing that speed without creating overwhelming incentives for individual companies or countries to defect.
If the risks are serious enough that the chief executive of a leading frontier laboratory believes capability development should deliberately slow, AI safety is no longer solely a technical problem.
It becomes a question of whether anyone can actually impose a speed limit on intelligence.
More to follow.
Sources: Reuters, 12 September 2026; Dario Amodei, We Must Pace the Frontier
Join the Dissenting Citizen Newsletter
Independent news and commentary delivered directly to your inbox.
