What Anthropic would not drop
Anthropic had been negotiating with the Department of War for months over the terms under which Anthropic would deploy Claude to the department and the intelligence community. The talks reached an impasse over two exceptions Anthropic asked to keep in place: mass domestic surveillance and fully autonomous weapons systems that act without human oversight. [1]
Amodei wrote that "AI-driven mass surveillance presents serious, novel risks to our fundamental liberties" and, on autonomous weapons, that "frontier AI systems are simply not reliable enough" to be deployed without a person responsible for the use of force. The statement frames Anthropic as otherwise willing: it says it supports deploying its models to the Department of War and the intelligence community, and that it has turned away hundreds of millions of dollars in revenue by restricting Chinese access to its models. [1]
Atlas interpretation: The department's stated position was that a vendor should not constrain how the military uses what it has bought. Read that way, this was not a dispute about whether Anthropic would sell to the government at all, only about which two uses it would sell for. [1]
What the department attached to the demand
Amodei's statement describes three threats the department made if Anthropic kept the safeguards: removal from its systems, a formal designation as a "supply chain risk," a label the statement says was until then reserved for US adversaries and never applied to an American company, and invoking the Defense Production Act to force the safeguards' removal. Amodei wrote that "these threats do not change our position: we cannot in good conscience accede to their request," and said Anthropic would help the department transition to another AI provider if it came to that. [1]
The next day, Secretary of War Pete Hegseth said he was directing the department to designate Anthropic a supply chain risk. Anthropic's response statement said the designation would restrict Claude only on Department of War contracts, not on its commercial customers or its work with other agencies, and repeated that it supports "all lawful uses of AI for national security" outside the two carve-outs. [2]
Atlas interpretation: The Defense Production Act is ordinarily invoked to compel production of something the government needs, not to compel a vendor to drop a usage restriction. Naming it here signals the scale of leverage the department was willing to use, since the Act's compulsory powers go well beyond an ordinary contract dispute. [1]
Sources
- Statement from Dario Amodei on our discussions with the Department of War
Anthropic · Feb 26, 2026
- Statement on the comments from Secretary of War Pete Hegseth
Anthropic · Feb 27, 2026