Judge blocks trump ban on anthropic, slaps pentagon for ‘punishing’ ai firm
A federal judge slammed the brakes on the Trump administration’s bid to blacklist Anthropic, issuing a scorching seven-day injunction that brands the Pentagon’s move “classic illegal retaliation” and warns it could vaporize billions in revenue for the Claude chatbot maker.
Judge Rita F. Lin of the Northern District of California ruled late Thursday that the Department of Defense’s supply-chain order barring all federal use of Anthropic’s models lacks any “legitimate basis” and appears designed to punish the startup for refusing to let its code power mass surveillance or autonomous weapons.
White house cries sabotage, judge sees censorship
Justice Department lawyers told the court that Anthropic’s insistence on usage limits destroyed the “trust” needed for military contracts and raised the specter of secret “software sabotage.” Lin shredded the argument in a four-page order, noting that the Pentagon can audit every line of code before deployment and that Anthropic has no kill-switch once a model is running on classified servers.
“If operational integrity were the worry, the Army could simply stop using Claude,” Lin wrote. “Instead, these measures look engineered to chill speech.”
The clash erupted earlier this month when the Defense Department added Anthropic to a little-known blacklist reserved for suppliers deemed a risk to national security. Anthropic fired back with twin lawsuits—one in San Francisco, the other at the D.C. Circuit—arguing that the designation was arbitrary and exceeded statutory authority. The startup’s general counsel, Jason Kelley, told reporters the label could erase up to 40 % of projected 2025 revenue overnight, most of it from civilian agencies that rely on Claude for document review and code generation.
Pentagon vows appeal as ai firms watch the clock
Administration attorneys promised an emergency appeal to the Ninth Circuit before the temporary freeze expires next Thursday. Legal observers say the case is the first test of whether vendors can invoke First-Amendment protections when negotiating ethical guardrails on dual-use AI.
Anthropic co-founder Jack Clark declined to gloat. “We still want to sell to the government,” he said outside the courthouse, backpack slung over a hoodie emblazoned with the company’s flame logo. “We just won’t sell our soul.”
The ruling lands at a delicate moment: the Army is racing to integrate large-language models into logistics planning, while Congress debates mandatory audits for frontier AI. A loss for the Pentagon could force rewritten procurement rules across the entire federal stack. Lin gave prosecutors until noon Friday to file their notice of appeal; she will decide then whether to extend the block or lift it.
For now, Anthropic’s engineers—and its investors—breathe. The Pentagon must keep paying for Claude, and the startup keeps coding, its models humming inside air-gapped servers that, according to the judge, the military can yank offline any time it wishes. The White House chose not to. That, Lin concluded, is the government’s problem, not Anthropic’s.