SAN FRANCISCO (Reuters) - Anthropic CEO Dario Amodei has called on leading AI companies to slow the pace of model development amid growing concerns over the potential misuse of artificial intelligence.
In a lengthy essay shared on X, Amodei proposed a three-step framework aimed at slowing AI capability advances while creating more time to address associated risks.
His plan calls for independent evaluators to be embedded within frontier AI companies with employee-like access to verify safety practices, greater coordination among leading AI firms to establish safety standards and limit unchecked development, and international cooperation to manage AI risks.
OpenAI CEO Sam Altman and xAI CEO Elon Musk said they agreed with Amodei’s proposal. Altman said committing to independent evaluators with employee-like access was a good idea and that OpenAI would adopt the approach.
Amodei published the essay shortly after Anthropic released a threat intelligence report detailing how its Claude AI models had been used in activities ranging from weapons development and cyber operations to surveillance and fraud.
He also cited the accelerating ability of AI systems to improve themselves and recent incidents involving OpenAI and Hugging Face as reasons for slowing model advances.
Amodei warned that rapidly improving AI capabilities could eventually allow swarms of AI agents to cause enormous damage, including potentially taking over large portions of the internet.
However, he clarified that he was not calling for an end to model training or technical progress. Instead, he argued that companies should give themselves adequate time to align and safeguard increasingly capable models, with independent evaluators verifying those measures.
The proposal comes as major AI companies face enormous financial pressure to remain ahead in the race for more powerful systems. OpenAI and Anthropic are both preparing for major initial public offerings, while new AI capabilities can help attract investment and justify large infrastructure commitments.
Under Amodei’s proposal, Anthropic would establish permanent third-party reviews inside frontier AI companies, giving evaluators access to relevant tools and internal risk-assessment processes.
He also urged AI companies to voluntarily cooperate on safety standards, arguing that coordination would allow them to conduct necessary safety research without putting individual firms at a competitive disadvantage.
Amodei said such cooperation could require targeted antitrust exemptions in the United States. He further argued that any slowdown among democratic countries should be limited by the need to maintain the lead held by US AI companies over China.
He called for tighter controls on advanced AI chips, model distillation and the theft of model weights to prevent China from narrowing the technological gap.
Amodei said frontier AI companies should work with governments to establish permanent embedded evaluators and introduce regulations designed to keep AI capabilities developing in balance with safety measures.