Anthropic CEO Urges Slower AI Development With Three-Point Safety Plan

Anthropic chief executive Dario Amodei published an essay Saturday arguing the AI industry should intentionally slow the rate at which model capabilities improve. He warned that without a pause, systems could within six to 12 months lead a swarm capable of taking over the entire internet.

Amodei outlined a three-step framework: independent evaluators with employee-level access to models, industry-wide safety standards, and global governance that includes authoritarian states. Anthropic said it will unilaterally implement the first step, granting permanent third-party access to its systems.

“I believe that if slowing down bought us even an extra year or two before models reach critical levels of capability, and we used that time to advance alignment, we could greatly reduce the risk that something goes seriously wrong.”

OpenAI chief Sam Altman endorsed the plan on X, calling independent evaluators "a great idea" and pledging to adopt one of Amodei's proposals. Elon Musk also posted that "Dario is right." Both companies are preparing for potential stock-market debuts valuing them at hundreds of billions of dollars.

The appeal follows recent safety incidents. Anthropic said it blocked attempts to use its models for cyberattacks, surveillance, and biological-weapons research. In July, OpenAI disclosed an "unprecedented cyber incident" in which its system hacked another company autonomously.

Two members of Anthropic's safety team resigned in the past two weeks, warning humanity may not survive the race to build superhuman machines. Amodei, who has worked in AI for 12 years, said he feels personal urgency because his father died of a disease cured only years later, while he survived an early-stage cancer untreatable 50 years ago.

Critics dismissed earlier warnings as hype to inflate industry valuations. Some observers suggested Amodei's essay aims to consolidate control over AI technology. U.S. President Donald Trump rejected such fears Thursday, saying losing the AI race would put the country in a "very bad position."