The Pentagon’s attempt to blacklist Anthropic has collapsed in court, exposing a broader struggle over who gets to set the limits of military artificial intelligence: elected officials and defense agencies, or the companies building the systems.

Judge Rita Lin ruled that the government’s sanctions against Anthropic were unlawful retaliation after the company refused to permit Claude, its AI model, to be used for domestic surveillance or autonomous weapons. She permanently barred enforcement of the sanctions and overturned the company’s designation as a “supply chain risk”—a label historically aimed at foreign threats. The ruling leaves the government free to appeal, but it immediately restores Anthropic’s position in federal contracting. ¹

Anthropic’s argument is fundamentally about constitutional limits. The company said the Pentagon punished it for publicly stated positions on AI safety and denied it due process before imposing the designation. Lin agreed, warning that “the empty invocation of national security is not a blank check to punish and retaliate against government critics.” ¹

The Pentagon presented the dispute differently. Officials argued that Anthropic’s refusal to accept military terms could create uncertainty during operations and potentially disable systems in the field. Their position was that the designation reflected a contractual and operational conflict—not retaliation over speech. That claim now stands against a ruling calling the designation “illegal and baseless.” ¹

The conservative framing emphasizes the constitutional stakes too, but from a sharper political angle: its headline describes the ruling as striking down Hegseth’s ban and condemning it as unconstitutional “retaliation.” ² The shared conclusion is that the Pentagon overreached; the unresolved question is whether military necessity can justify compelling private AI companies to accept uses they consider unsafe or unlawful.