Anthropic’s CEO says the AI race needs to slow down

Anthropic CEO Dario Amodei is urging AI companies to immediately slow the rate at which they make their most powerful models more capable, arguing that safety work is falling behind. He is not calling for AI to stop: he sees major potential benefits, but warns that rapid advances—partly because AI is helping create its own successors—could outstrip people’s ability to test and control the systems, while making cyberattacks and biological misuse easier. His proposed answer is to give safeguards and independent scrutiny time to catch up; Anthropic says it will give an outside safety-testing group, METR, unusually deep access inside the company. The shift is from simply racing to build the strongest model to proving that independent checks can keep up with it.