Anthropic faces Pentagon fallout, urges stronger AI oversight

Anthropic faces Pentagon fallout over AI use, calls for mandatory third-party testing, binding regulation, and new frameworks for government oversight and economic resilience.

Anthropic CEO Dario Amodei expressed uncertainty about whether the company’s AI model was involved in missile strikes against Iran, following a dispute with the US Department of Defense. This conflict led to Anthropic being banned from Pentagon systems after the Pentagon demanded unrestricted access to Anthropic’s Claude AI model, which had been used in military intelligence. Anthropic refused to allow its technology for mass surveillance or fully autonomous weapons, prompting the Pentagon to label it a supply-chain risk. Separately, Amodei advocated for mandatory third-party testing of advanced AI models, focusing on four risk categories, and announced a legislative proposal for independent evaluations. Anthropic also launched Mythos 5, supporting a standardized testing framework. In an essay, Amodei argued that transparency is no longer sufficient and called for binding regulation for powerful AI, modeled after FAA safety standards. Anthropic’s new policy frameworks propose government authority to block risky AI deployments, impose revenue-based penalties, and ensure public resilience, with a focus on advanced models and economic impacts for workers.