Anthropic has reopened negotiations with the Pentagon in an effort to preserve a $200 million AI pilot contract and avoid being pushed out of the U.S. defense supply chain. The dispute centers on how far Claude can be used in military settings: the government wants the model available for “any lawful use,” while Anthropic is holding to two specific limits.
Contract terms became the main fault line
Since last year, Anthropic has worked under a pilot agreement with the U.S. Department of Defense, making Claude the first advanced AI model cleared for deployment on classified networks. Talks broke down last week, and the Pentagon then labeled Anthropic a “supply chain risk.” The trigger was a push to revise the contract. The Trump administration insisted that the model be usable for “any lawful use,” without carve-outs.
Anthropic refused to remove its restrictions. The company’s two red lines are clear: Claude cannot be used for mass domestic surveillance of U.S. citizens, and it cannot be used in fully autonomous lethal weapons, meaning systems that can select and attack targets without human involvement. Anthropic’s position is that such applications could threaten democratic values, and that current AI systems are not reliable enough for those roles.
Amodei is now negotiating directly
According to people familiar with the matter, CEO Dario Amodei is now in direct talks with Emil Michael, the Pentagon’s undersecretary of defense for research and engineering. The goal is to reach a compromise that keeps Claude in use by the military while sharply reducing the risk that Anthropic is formally blacklisted.
The pressure has been building for days. Defense Secretary Pete Hegseth had issued an ultimatum, demanding that Anthropic make concessions by a deadline or face contract termination and possible enforcement under applicable rules. After talks collapsed on Friday, this renewed contact suggests both sides still see room for a negotiated outcome.
The fight is really about limits on military AI
If a new deal is reached, it could ease tensions and shape how AI companies work with the U.S. defense establishment. The report notes that OpenAI has already struck an agreement with the Pentagon and is also adjusting contract language to include similar limits. Anthropic’s message has stayed consistent: it is willing to cooperate, but not at the cost of its core safeguards. If no agreement is reached, the company says it will help ensure a smooth transition to another supplier so military operations are not disrupted.
The episode captures a wider conflict that has been building across the AI sector. Companies talk about ethical and safety boundaries; the Pentagon is focused on battlefield flexibility and technological advantage. Anthropic has built its identity around responsible AI, and in this case it is not dropping its conditions. Whether the talks succeed now depends on whether those two red lines can remain in the final contract.