A federal judge on Thursday said the Trump administration has not produced enough evidence to justify labeling Anthropic a supply-chain risk or banning the federal government from buying its AI. US District Judge Rita Lin made the finding during a hearing in one of two lawsuits Anthropic filed against the Department of Defense in March 2026, challenging both the ban and the risk designation. Lin had already temporarily blocked the ban in March and is now weighing whether to make that order permanent.
The dispute traces to stalled contract negotiations between Anthropic and the DOD. Anthropic said it did not want its AI used for mass surveillance of Americans or for targeting or firing decisions involving lethal weapons, arguing the technology was not ready. The Pentagon countered that a private company should not dictate how the military uses technologies, and said it would deploy the tools in lawful ways.
The government went further, arguing that Anthropic's public criticism of the DOD itself justified the ban. Lin flagged that logic as a serious problem, warning it could set a precedent of retaliating against federal contractors who disagree with the administration in public.
Key facts
- 01US District Judge Rita Lin said Thursday the Trump administration has not shown enough evidence to label Anthropic a supply-chain risk or ban federal use of its AI.
- 02The dispute stems from stalled Anthropic–DOD contract talks over use of the company's models for mass surveillance and lethal-targeting decisions.
- 03Lin said she saw no proof Anthropic could alter a delivered model or 'flip some kind of kill switch,' rejecting a core DOD claim.
- 04Anthropic filed two lawsuits against the DOD in March 2026 challenging the ban; a second case is being heard in Washington.
- 05Lin temporarily blocked the ban in March and is now weighing whether to make the order permanent.
The judge described the retaliation theory in blunt terms from the bench.
The DOD also claimed Anthropic could potentially disable or alter its AI models during warfighting operations, a technical assertion that outside researchers have said lacks supporting evidence. Lin agreed on the record, saying she saw no proof Anthropic could alter a delivered model or flip some kind of kill switch. That finding matters because the kill-switch claim has been the centerpiece of the DOD's national-security rationale for the supply-chain risk label.
Anthropic's position throughout the case has been narrow rather than sweeping. The company has not refused all defense work; it has objected to specific uses — mass surveillance of Americans and autonomous lethal-targeting decisions — while continuing to sell to other parts of the federal government. That distinction is what makes the retaliation question load-bearing: if a vendor can be blacklisted for objecting to particular use cases, the practical effect is to force acceptance of every use case as a condition of doing business with the government.
The DOD, for its part, has framed the fight as a matter of who controls military technology decisions once a contract is signed. Its argument is that a commercial AI vendor cannot condition delivery on approval rights over downstream uses, particularly in classified or operational settings. That principle is defensible on its own terms, but Lin's ruling suggests it does not extend to punishing a vendor before any contract exists.
The second Anthropic suit is being heard separately in Washington. Between them, the two cases will produce the first substantive judicial record on how the federal government can and cannot use procurement designations against frontier AI vendors that push back on specific deployments. A permanent injunction from Lin would sharply narrow the DOD's leverage in future negotiations with Anthropic and, by extension, with every other lab weighing similar red lines.
The counterweight worth naming is that Thursday's hearing was preliminary in posture — Lin has not yet issued a final order, and the government can still supplement the record or appeal an adverse ruling. The DOD may also refine its technical claims about model modification, though it would need evidence it did not bring to this hearing. And the parallel Washington case could reach a different conclusion on overlapping questions.
For the broader AI market, the significance sits in the precedent. Frontier labs have been publishing use-case restrictions — no mass surveillance, no autonomous lethal targeting, no election manipulation — for years, and selling into government at the same time. If a court rules that a supply-chain risk label cannot be used to punish those restrictions, labs gain real leverage to hold their published policies without losing federal revenue. If the DOD prevails on appeal, the opposite: published policies become negotiating positions the government can override by threatening designation. Anthropic is the test case, but the outcome will shape how every lab, including OpenAI and Microsoft's federal-facing units, writes its next defense contract.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
