Anthropic CEO urges government to block unsafe AI models

The chief executive of Anthropic has called for the U.S. government to be granted the power to block the deployment of advanced artificial intelligence systems deemed to pose unacceptable risks, advocating for a mandatory safety regime modeled on aviation regulation. Dario Amodei made the case in a lengthy essay published on June 10, arguing that the rapid pace of AI development has rendered voluntary transparency measures insufficient.

Amodei proposed a regulatory framework where frontier AI models exceeding a specified computational threshold would be subject to mandatory third-party audits before public release. These audits would assess risks across four categories: cybersecurity vulnerabilities, biological weapons capabilities, the potential for models to operate beyond human control, and the ability to accelerate automated research in dangerous domains. Should a model fail these safety evaluations, the government would have the authority to prevent its deployment.

He explicitly compared the necessary oversight to that governing airplanes, automobiles, and pharmaceuticals, describing AI as a powerful technology essential to the modern economy but capable of causing...