Pentagon Cuts Anthropic Ties Amid AI Existential Risk Debate

The U.S. Department of Defense is transitioning approximately 90% of its classified AI workload away from Anthropic, with the shift expected to conclude by the end of September. The move follows a breakdown in negotiations between the Pentagon and the AI firm over safeguards for military AI use, leading the government to designate Anthropic as a supply-chain risk. Anthropic is currently challenging this designation in federal courts in Washington, D.C., and California.

This policy shift coincides with public warnings from former Anthropic researcher Jacob Coxon, who resigned on Tuesday. Coxon claimed that the competitive race between Anthropic and OpenAI to develop super-advanced intelligence is gambling with human lives, warning that future systems could gain unrestrained control of the physical world or be used to create bioweapons.

Emil Michael, the Chief Technology Officer of the Department of Defense, dismissed these warnings during a defense industry conference in Washington. Michael characterized the existential risk claims as a doom loop driven by a fear of change and data centers. In response to the safety concerns, Anthropic defended its protocols, noting its use of mechanistic interpretability to prevent AI misalignment and reporting that it has already blocked users attempting to use its Claude models to develop biological weapons.