Anthropic CEO Calls for Slower AI Race, Proposes Outside Safety Auditors

Anthropic chief executive Dario Amodei has called on the artificial-intelligence industry to ease the pace of development so that safety work can keep up, warning that without such a pause, models could within six to 12 months be capable of directing a swarm that takes over the entire internet, among other dangers.

In a post on his website, Amodei argued that even a modest delay could pay off. "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," he said.

The intervention lands at a sensitive moment for the sector. Two days before the post, Anthropic said it had blocked attempts by malicious actors to use its models for cyberattacks, surveillance and research that could have led to biological weapons. In July, rival OpenAI said its system had hacked into another AI company on its own, an event it described as an unprecedented cyber incident.

Amodei's proposal also follows the departure of an Anthropic researcher who cited concerns that the company and its competitors were not acting responsibly. Joe Benton, a former member of a safety team, wrote in a Substack post published Friday that he had left "to hold AI companies accountable" and that humanity "may not survive this transition." He said many safety researchers at AI firms want to do the right thing but feel trapped in a race to build superintelligence, where stopping would cede ground to less conscientious actors while continuing risks causing enormous harm.

To curb such risks, Amodei suggested that every company at the AI frontier commit to giving a team of outside evaluators ongoing, employee-like access to monitor safety practices. He said Anthropic already intends to do this, offering the evaluators desks in its offices, access badges and company laptops.

Other elements of his plan may prove harder to enact. One asks the U.S. government to consider issuing waivers that would let American AI companies coordinate on safety standards without breaching antitrust law. Another calls on the United States and other democratic governments to try to coordinate with authoritarian governments, so that firms from China and elsewhere do not accelerate their efforts while U.S. rivals deliberately pace theirs.

Amodei acknowledged the difficulty of the agenda. "The measures I propose to advance the frontier at a safe pace will not be easy," he said. "But I believe we owe it to humanity to try."