A U.S. federal judge has permanently blocked the Pentagon from enforcing a supply-chain risk designation against Anthropic, ruling that the Department of Defense acted illegally when it blacklisted the Claude maker over the company's public safety positions on battlefield AI.
What the court said
In a 59-page order issued Thursday night, U.S. District Judge Rita Lin of the Northern District of California found that Defense Secretary Pete Hegseth's designation violated Anthropic's First Amendment rights and its Fifth Amendment due-process protections. Lin ordered the government to rescind all directives issued against Anthropic and imposed a permanent injunction. "The empty invocation of national security is not a blank check to punish and retaliate against government critics," she wrote.
How the fight started
The dispute grew out of a roughly $200 million Pentagon contract over deploying Anthropic's Claude models on classified systems, in which Anthropic sought contractual limits blocking use of its models in autonomous lethal weapons or domestic mass surveillance. The Defense Department rejected those limits and Hegseth then designated the company a supply-chain risk — the first time a U.S. company has been publicly labelled that way under an obscure procurement statute typically used to guard against foreign sabotage. Executives said the ban risked costing Anthropic billions of dollars in future business.
Why it matters for defense-AI vendors
The ruling is a landmark moment for AI vendors negotiating usage rules on classified programs, and clarifies how far the executive branch can go in punishing contractors whose safety positions differ from government policy. Anthropic said it remains "focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology"; the Pentagon has said nothing publicly, and appeals are widely expected.
The ruling lands the same week Anthropic joined OpenAI and 100+ companies in an open letter warning about AI-enabled cyberattacks — an unusual sequence of policy fights for a five-year-old lab now central to U.S. AI governance debates.
Reporting based on coverage from Al Jazeera / AP / Reuters (Aug. 28, 2026), NOTUS, Axios and The Hill.