Anthropic chief executive Dario Amodei said the company would not remove two safeguards from its agreements with the US Department of War, despite what he described as threats to end the department's use of Claude and designate Anthropic a supply-chain risk. His statement cast the dispute as a narrow disagreement over limits rather than opposition to military adoption of artificial intelligence.
Amodei said Anthropic regards AI support for the United States and other democracies as important and has actively deployed its technology for national-security work. According to the company, Claude is used across the department and other agencies for intelligence analysis, modeling and simulation, operational planning and cyber operations. Anthropic also said it was the first frontier AI provider to place models on classified US government networks and at National Laboratories. These claims were presented by Anthropic and were not independently corroborated in the supplied material.
The company said it recognizes that military decisions belong to the government, not a contractor, and that it has not objected to individual operations or tried to impose case-by-case restrictions. However, Amodei argued that a narrow category of uses could undermine democratic values or exceed what current AI systems can perform safely and reliably. He said two related exceptions had long been absent from the company's contracts and had not previously blocked wider adoption. The supplied statement excerpt did not specify the exact wording of those exceptions.
Amodei said the department was now requiring AI suppliers to permit “any lawful use” and remove the safeguards. He alleged that officials had raised three possible responses: taking Anthropic models out of government systems, applying a supply-chain-risk designation and using the Defense Production Act to compel removal of the restrictions. No department response was included in the evidence packet.
Anthropic argued that labeling it a risk while treating its product as essential enough for compelled access would be contradictory. Even so, Amodei acknowledged the department's authority to choose contractors whose policies match its requirements. He said the company preferred to continue supporting defense users under the terms it had proposed.
If officials chose another supplier, Anthropic would help manage a transition to avoid disruption to military planning, operations and other critical missions, the statement said. The company pledged to keep its models available on its proposed terms for as long as needed during that process.
The statement also emphasized Anthropic's broader security posture. It said the company had forgone several hundred million dollars by cutting access for firms linked to the Chinese Communist Party, disrupted attempted state-sponsored cyber abuse of Claude and supported stronger chip export controls. Those assertions reinforce Anthropic's central position: it sees itself as a willing national-security partner, but not one prepared to abandon the safeguards at issue.
