Federal Court Strikes Down Pentagon's Supply-Chain Risk Label for Anthropic

First Amendment Protections in the AI Procurement Arena

A federal court decision late Thursday has set a precedent that could reshape how the U.S. government engages with AI vendors who push back on defense applications. U.S. District Judge Rita Lin ruled that Defense Secretary Pete Hegseth's designation of Anthropic as a national security risk violated the company's First Amendment rights and denied it due process under the Fifth Amendment. The ruling described the government's actions as "arbitrary and capricious," language that signals a fundamental failure in administrative procedure.

The decision arrives at a moment when the boundaries between commercial AI development and national security applications remain poorly defined. At DailyTechWire, we've tracked similar tensions in Seoul, where Naver faced pressure over its HyperCLOVA X model's military applications, and in Singapore, where government procurement guidelines now explicitly address vendor refusal rights. The California ruling suggests American courts may impose harder limits on executive branch discretion than many observers expected.

The Guardrails Dispute That Triggered Federal Action

The conflict began when Anthropic established firm restrictions on how its Claude models could be deployed by government customers. The company refused to permit use cases involving fully autonomous weapons systems and mass surveillance of U.S. citizens, two categories that have become flashpoints in the broader debate over AI governance. Defense officials countered that they intended only lawful applications and argued that Anthropic was attempting to control technology after sale, a position that raised questions about ownership and vendor oversight in sensitive procurement.

Earlier this year, the administration escalated by ordering all federal agencies, including those outside the defense sector, to cease working with Anthropic. The breadth of that directive was unusual. Typically, supply-chain risk designations focus narrowly on specific contracts or agencies where exposure is highest. The government-wide ban suggested punitive intent rather than targeted risk management.

Judge Lin's opinion highlighted internal contradictions in the government's position. While Hegseth labeled Anthropic a threat, he simultaneously proposed invoking the Defense Production Act to compel the company's cooperation, a move that treats a firm as essential to national security rather than dangerous to it. The Department of Defense also continued pursuing contracts with Anthropic and collaborated with the company on Mythos, a new model designed for cybersecurity applications. Those actions undercut the rationale for the supply-chain designation.

What the Ruling Actually Says About Retaliation

Lin's decision rested heavily on evidence that the government sought to "make a public example" of Anthropic for what officials described as the company's "arrogance" in criticizing procurement practices. That language, drawn from internal communications and public statements, provided the factual basis for the First Amendment retaliation claim. The court found that the designation was not a good-faith assessment of supply-chain vulnerability but a response to protected speech.

The ruling also addressed technical claims about backdoor access. The government had suggested that Anthropic could retain control over deployed models, enabling the company to influence or disable systems in use. Lin found that Anthropic "undisputedly lacks" such access once technology transfers to the Department of Defense. That factual determination mattered because it eliminated the plausible security rationale for the designation. Without a credible technical threat, the government's actions appeared to rest entirely on policy disagreement.

The court was explicit about the limits of national security invocations. Lin wrote that "the empty invocation of national security is not a blank check to punish and retaliate against government critics." That sentence will likely be cited in future cases where vendors challenge procurement exclusions. It establishes that courts will scrutinize the factual basis for such designations rather than defer automatically to executive assertions.

Implications for AI Vendors and Defense Procurement

Anthropic filed two separate complaints in March, one in California and another in Washington, D.C. The California case produced Thursday's ruling; the D.C. litigation remains pending. The dual-track strategy is common in administrative law challenges, where plaintiffs seek the most favorable venue while preserving alternative paths. The D.C. Circuit has deep expertise in national security cases, and its eventual ruling may carry more weight with other agencies even if the substantive outcome aligns with Lin's decision.

For AI companies navigating defense contracts, the ruling clarifies that setting use-case restrictions does not automatically invite retaliation under the guise of supply-chain risk. That protection matters most for firms with commercial business models that depend on public trust. Anthropic, like OpenAI and several smaller model developers, markets itself partly on safety commitments. A government label implying security threats could erode consumer and enterprise confidence, creating spillover harm beyond lost federal revenue.

The decision also exposes a tension in how Washington thinks about strategic industries. If the U.S. government believes frontier AI is critical to national competitiveness and security, as policy documents and funding priorities suggest, then treating leading developers as adversaries when they resist specific applications seems counterproductive. Other democracies have adopted co-development frameworks that negotiate acceptable use cases upfront rather than imposing them unilaterally. The ruling may accelerate interest in similar models domestically.

What Comes Next for Anthropic and the Pentagon

Anthropic's public statement welcomed the ruling and emphasized its willingness to "work productively with the government to harness AI for our national security." That language signals a preference for settlement over prolonged litigation. The company has invested heavily in government relations since its founding, hiring former defense and intelligence officials to bridge cultural gaps between Silicon Valley and Washington. A negotiated framework that defines permissible and prohibited use cases could serve both sides better than continued courtroom conflict.

The Department of Defense has not yet commented on whether it will appeal. An appeal would go to the Ninth Circuit, which has been less deferential to executive authority on First Amendment grounds than some other appellate courts. The administration faces a strategic choice: accept the loss and negotiate new terms, or risk a circuit-level opinion that further constrains its flexibility across the AI vendor landscape.

The case also intersects with broader regulatory efforts. The Commerce Department is developing supply-chain security rules for AI infrastructure under export control statutes, and the Cybersecurity and Infrastructure Security Agency has proposed vendor assurance requirements for critical software. If courts impose strict scrutiny on retaliation claims, agencies will need clearer criteria and more rigorous documentation before excluding vendors. That procedural burden could slow designations but improve their durability.

For now, the ruling removes the formal supply-chain risk label, but it does not compel the Pentagon to contract with Anthropic. The government remains free to choose other vendors, as Lin acknowledged. What it cannot do, at least under this decision, is use procurement exclusions as punishment for advocacy on safety policy. That distinction will matter as more AI companies weigh whether to accept or refuse controversial government applications in the years ahead.