Pentagon slaps 'supply-chain risk' label on anthropic, freezing claude out of federal deals
The Biden administration wants a federal judge to keep Anthropic on the Pentagon’s blacklist, arguing the AI startup’s refusal to let the military do “any lawful thing” with its models turns Claude into a battlefield wildcard worth billions in lost contracts.
The no-show clause that broke the camel’s back
Inside a terse 18-page filing in San Francisco late Tuesday, Defense Department lawyers said every agency that tried to license Claude hit the same wall: Anthropic balked at striking the standard “government-purpose rights” language that lets soldiers repurpose code on the fly. When generals asked for a carve-out for autonomous weapons or mass-surveillance drills, the company countered with hard nos rooted in its “constitutional” AI safety policy. The stand-off escalated until the Pentagon’s acquisition office stamped the startup “supply-chain risk,” a bureaucratic death sentence that bars it from new federal money.
The label isn’t ceremonial. Since February, any program manager who even cites Claude in a slide deck must file a 15-page waiver. None has been approved. Anthropic claims the freeze cost it $3.2 billion in forecast revenue and forced layoffs inside its D.C. policy shop. A company spokesperson told TechBloom the move “weaponizes procurement rules to punish speech,” teeing up a First-Amendment clash that lands in court on 24 March.
Government: ‘we’re not buying a conscience, we’re buying code’
Justice attorneys counter that the fight is plain-vendor shopping, not censorship. They quote an internal email from an Air Force colonel: “If we can’t patch it at 3 a.m. during a surge, it’s not software, it’s a pet.” The filing also flags Anthropic’s update cadence—weekly tweaks shipped from San Francisco—as a potential kill switch the firm could flip at any time. “A contractor that retains root access to the brain of a targeting algorithm is itself a threat vector,” the brief warns.
Dario Amodei, Anthropic’s CEO, fired back in a blog post: “We never asked to vet missions; we only asked not to build killing machines.” The distinction, the government says, is meaningless once bullets fly. Legal scholars note the case could cement whether “ethical AI” pledges are enforceable commercial terms or mere marketing gloss.
What hangs in the balance
Beyond the balance sheet, the ruling will shape how the U.S. stocks its next-gen warfare toolkit. If Anthropic wins, every cloud vendor will rush to insert conscience clauses, fragmenting the military’s AI stack into tiers of ethical clearance. If the Pentagon prevails, startups must choose between principled posturing and the biggest check in tech. Either way, the first gavel drop on 24 March will echo louder than any boardroom pitch deck.