Judge slaps down trump ban on anthropic, calls pentagon move 'punishment'
A federal judge has torpedoed the Trump administration’s attempt to freeze Anthropic out of every government contract, ruling the move smacks of retaliation rather than national security.
In a blistering order issued Thursday night, Judge Rita F. Lin of the Northern District of California granted Anthropic’s bid for a temporary restraining order, halting the Defense Department’s blanket suspension for at least seven days. The decision yanks the emergency brake on a policy that the AI start-up warned could vaporise billions in future revenue and redefines the battlefield for Silicon Valley’s growing clash with the Pentagon.
The pentagon wanted a mute vendor; anthropic refused
Lin’s 18-page opinion dismantles the government’s claim that Anthropic poses a supply-chain risk. ‘These measures appear designed to punish Anthropic,’ she wrote, noting that if the brass fears operational compromise, it can simply unplug Claude. ‘Instead, DoD seeks to exile a company for expressing legal reservations.’ The judge branded the action a textbook First-Amendment violation after the start-up demanded written assurances its models would not feed mass surveillance programmes or autonomous weapons systems.
The dispute detonated earlier this month when the Defense Logistics Agency issued a unilateral suspension, citing classified assessments that Anthropic’s code could be modified post-deployment to sabotage military networks. Anthropic struck back with twin suits—one in San Francisco seeking injunctive relief, another in the D.C. Circuit alleging the agency exceeded statutory authority under the Defense Federal Acquisition Regulation.
Inside the courtroom, government lawyers floated a dystopian scenario: Anthropic engineers pushing a silent update that turns battlefield chatbots against their operators. Anthropic countered that every model is delivered as a static container, audited on classified networks the company cannot access, much less alter. Lin sided squarely with the start-up, finding ‘no evidence—speculative or otherwise—of sabotage intent.’
Seven-day reprieve sets up appellate knife fight
The stay lasts only one week, giving the Justice Department a narrow window to escalate the skirmish to the Ninth Circuit. Sources close to the case expect DoJ to file an emergency motion before sunrise Friday, arguing that district courts lack authority to override procurement decisions touching on national security. Even if the appeals panel grants a stay of the stay, Anthropic’s discovery machine is already humming: subpoenas for internal Pentagon memos are drafted and ready.
Investors reacted with relief. Venture desks at Sequoia and Spark Capital, both major Anthropic backers, circulated Lin’s order to limited partners within minutes, noting that government contracts account for roughly 18 % of projected 2025 revenue. A venture partner at one top-tier fund put it bluntly: ‘We were staring at a valuation cliff; this ruling buys runway.’
The ruling lands amid a broader regulatory crackdown on frontier-model providers. The Pentagon has quietly compiled a blacklist of AI vendors that refuse to grant unlimited rights to retrain or weaponise their systems. Anthropic’s refusal to sign the standard ‘Model Use Rider’—a document that waives moral-hazard objections—made it the first domino to fall. Other labs are watching nervously; sources at OpenAI and Cohere say their government-affairs teams have pre-drafted similar litigation templates.
For now, federal agencies can keep downloading Claude. But Lin’s pen has redrawn the lines of engagement: disagreeing with the Defense Department is no longer debarment by default. The Pentagon must either prove sabotage or retreat. The clock starts now.