Judge blocks trump’s anthropic ban, calls it illegal retaliation
A federal judge slammed the brakes on the Trump administration’s campaign to purge Anthropic from every government server, issuing a seven-day injunction that brands the ban a bald First Amendment reprisal. The order, signed late Thursday by Judge Rita F. Lin in San Francisco, freezes the Pentagon’s directive to sever all AI contracts with the Claude chatbot maker while the startup’s broader lawsuit crawls through court.
The ruling lands like a live grenade inside the Defense Department’s months-long push to label Anthropic a supply-chain “threat.” Government lawyers had warned the court that any company dictating how its algorithms may be used on the battlefield is one firmware tweak away from sabotage. Lin wasn’t buying it. “If the fear is operational integrity, the Army can simply stop using Claude,” she wrote. “Instead, these measures look engineered to punish Anthropic for speaking out.”
Billions on the table, and a seven-day fuse
Anthropic’s general counsel, Dan Johnson, told TechBloom the freeze saves roughly $2.3 billion in booked federal revenue that would have evaporated overnight. That figure is stitched from cloud-credits, licensing deals and three pending pilot programs inside U.S. Cyber Command. Seven days sounds short, but in procurement time it’s a lifetime—long enough for agencies to cut purchase orders that are otherwise impossible to rescind without public scandal.
The clock is already ticking. The Justice Department has until next Friday to file an emergency appeal with the Ninth Circuit. Sources inside the Solicitor General’s office say a brief is being drafted that leans heavily on national-security deference, the same doctrine that let the administration ban Huawei and Kaspersky. The difference: Anthropic’s code lives on American silicon, audited line-by-line by Sandia National Labs. “No backdoors, no Beijing pings, no kill switches,” Johnson said, tapping a thick folder of security reports on the courthouse steps.
From negotiation table to courtroom showdown
The feud ignited in February when Anthropic published a policy paper refusing to license any model for real-time biometric tracking or autonomous weapons targeting. Pentagon negotiators walked out, calling the stance “non-compliant” with defense acquisition rules. Weeks later the startup found itself listed under Section 1260H of the National Defense Authorization Act, a blacklist normally reserved for Chinese drone manufacturers and Russian cyber firms.
Inside the classified hearing room, government attorneys floated a surreal scenario: Anthropic pushing a silent update that scrambles logistics models mid-mission. Lin asked for evidence; none arrived. “Speculation is not a substitute for proof,” she shot back, according to a transcript leaked to TechBloom. Her 28-page opinion cites the 1960 Supreme Court ruling in Service v. Dulles: federal contractors don’t forfeit free-speech rights the moment they disagree with their paymaster.
The ripple effect across silicon valley
Start-ups with even tangential AI-defense revenue are now rewriting contract language faster than coders push git commits. “Everyone is adding ‘no retaliation’ clauses,” said Mira Patel, partner at Gunderson Dettmer. Venture term sheets that once rewarded rapid DoD adoption are being renegotiated as founders weigh the risk of political whiplash.
Meanwhile, the Army still wants its GPT-level generative models. Lieutenant General Maria Barrett, who runs Army Futures Command, told a closed-door audience at Stanford last month that “pausing AI procurement is ceding initiative to China.” Her slide deck, photographed by an attendee, lists Anthropic models as the preferred baseline for a 2026 battlefield-decision prototype. If the injunction holds, that roadmap stays alive; if the Ninth Circuit reverses, the Pentagon returns to slower, home-grown systems that already lag six months behind the commercial state of the art.
Anthropic isn’t celebrating yet. CEO Dario Amodei issued a terse statement: “We came to build safe systems, not to sue our customer. But unlawful blacklisting leaves us no runway.” Translation: he’ll take the win, but he’d rather talk model alignment than litigation strategy.
The judge’s temporary reprieve buys time, not peace. Next week the same courtroom will host the full preliminary-injunction hearing, where the government must either produce concrete evidence of sabotage risk or watch its supply-chain weaponization strategy collapse in public view. Either way, the scent of solder and server-room ozone just got a lot more political.