Anthropic arrives at San Francisco federal court today, seeking an emergency injunction to halt the Pentagon's blacklist of its Claude models. The startup argues the Department of Defense's "supply chain risk" designation threatens billions in revenue and stalls critical government integrations.

Judge Rita Lin presides over the hearing, where the core technical question remains: does Anthropic retain enough control over deployed models to enable sabotage? The DoD flagged the company in March, marking the first time a U.S. AI developer faced such a national security label. Consequently, major contractors like Microsoft and Amazon must certify they aren't running Claude in defense projects. Palantir, however, confirmed continued usage despite the controversy, noting models remain active in Iran operations.

Tensions escalated after negotiations stalled on the GenAI.mil platform last September. Anthropic refused clauses allowing fully autonomous weapons or domestic mass surveillance, terms the Pentagon insists it doesn't require. In February, President Trump ordered agencies to cease using the technology, citing ideological conflicts on social media. This followed a $200 million agreement signed the previous July to deploy technology across classified networks.

Anthropic's legal team contends the ban inflicts irreversible reputational damage without evidence of actual risk. They emphasize that an injunction wouldn't force the government to keep using Claude, merely allow business to continue during litigation. Engineers watching closely note the precedent this sets for model deployment in classified environments. If the designation stands, vendor certification processes will tighten significantly. Lin may issue a ruling from the bench or return with a written decision later this week. For the ML industry, the outcome defines the boundaries between national security protocols and commercial AI deployment.

Source: CNBC