The Pentagon’s dispute with Anthropic over permitted military uses of the Claude AI system has escalated beyond a contract negotiation. Anthropic said Feb. 27 that Defense Secretary Pete Hegseth had directed the department to designate the company a supply-chain risk after it refused to remove restrictions involving mass domestic surveillance of Americans and fully autonomous weapons.
The action, if formally implemented, could affect a $200 million Pentagon relationship and put defense contractors using Claude into an unsettled compliance position. It also turns a commercial disagreement into an early test of whether a frontier-AI developer can retain enforceable limits on government uses after selling technology to the U.S. military.
There is an important distinction in the public record: Anthropic reported a direction to pursue a designation, not publication of a final designation or order. The company said it would challenge any such action in court. Separately, reported Pentagon consideration of the Defense Production Act was an option or threat under discussion, not a confirmed invocation of the law.
Deadline led to a wider confrontation
The immediate clash followed days of negotiations. ABC News reported that Pentagon officials had given Anthropic a Friday deadline to accept broader military-use terms or risk losing Pentagon business and facing additional measures. The outlet said Anthropic was participating in a Pentagon pilot program valued at $200 million.
CBS News reported that officials sent the company what sources characterized as a best-and-final offer, with a 5 p.m. Eastern deadline on Friday. CBS also reported that Anthropic received a $200 million Pentagon contract in July. Those figures describe a meaningful but still limited direct commercial relationship; the larger issue is whether restrictions could constrain other defense-sector deployments of the model.
Anthropic’s position is not that Claude should be unavailable to national-security agencies. In its public statement, the company said it supports lawful national-security uses apart from the two stated exceptions: mass domestic surveillance and fully autonomous weapons. It said, to the best of its knowledge, its policies had not interfered with a government mission.
Pentagon officials gave a different account of the practical risk. They said the department sought lawful use of the technology, and that the end user is responsible for ensuring legality. Officials also said the Pentagon did not intend to employ the tools for domestic surveillance or autonomous weapons. The disagreement is therefore less about a disclosed operational plan than about the contractual mechanism that would control future uses and whether Anthropic’s safety conditions could be overridden.
Competing readings of the proposed safeguards
Anthropic said the latest proposed language made virtually no progress toward its two restrictions and could enable its safeguards to be disregarded. The company’s account does not establish that the Pentagon plans either prohibited use. Rather, it argues that reliance on an end user’s legal obligations is insufficient when its system could be directed into sensitive intelligence or weapons workflows.
The Pentagon’s formulation places the operative compliance burden on the agency or contractor deploying the model. Anthropic wants the use boundaries embedded in the terms under which its technology is made available. That difference has commercial consequences: a supplier-level restriction can apply before a model is integrated into a workflow, while an end-user legality standard can leave the supplier with less say over a deployment once access is granted.
The dispute comes as contractors and federal agencies are testing generative AI for tasks ranging from analysis and coding to administrative support. A contract term that determines which party bears responsibility for screening a use case can shape adoption as much as model performance or price. For defense vendors, a government directive concerning a model provider could create procurement and legal-review questions even before a court resolves the underlying authority.
What a supply-chain designation could cover remains unresolved
Anthropic said Hegseth had suggested that companies doing business with the military could be restricted from doing business with Anthropic. But the company contends that a designation under 10 U.S.C. 3252 could apply only to Claude use in Department of War contract work, not to commercial customers or to contractors’ work outside department contracts. That is Anthropic’s legal interpretation, not a settled description of the scope of a final action.
The statutory question is central because the economic effect differs sharply between those alternatives. A restriction confined to Department of War contract work would target deployments connected to that customer. A restriction reaching every company that does military business while also buying Anthropic services would have a far broader spillover into the defense industrial base. No final published designation in the available accounts resolves which approach the department would take.
Anthropic called the threatened action unprecedented against a U.S. company. Geoffrey Gertz of the Center for a New American Security told NPR’s reporting, republished by Houston Public Media, that such tools have traditionally been aimed at technology tied to foreign adversaries. That comparison helps explain why the designation threat has drawn attention: it would deploy a supply-chain security framework in a conflict with a domestic supplier whose product the Pentagon has itself been procuring.
Reporting also raised the prospect of using the Defense Production Act to compel access. A Pentagon official described that possibility to ABC News, while CBS reported that officials were considering it. Neither account said the government had invoked the act, and neither establishes that the statute would ultimately provide the authority sought. Any move in that direction would likely produce a separate legal fight over the government’s power to compel service on terms the supplier rejects.
For now, the sequence is clearer than the outcome. Reports on Feb. 25 and 26 described the deadline and the final offer; by Feb. 27, Anthropic had publicly said Hegseth directed a supply-chain-risk designation and that it would contest one in court. The next consequential document is likely to be a formal department action defining both the restricted technology and the entities, contracts and uses it would cover.