Summary: On March 26, 2026, Judge Rita Lin issued a preliminary injunction blocking the Pentagon’s supply-chain-risk designation of Anthropic, calling it likely illegal First Amendment retaliation. The order restores the situation before the February 27 directives after one week. The Pentagon may still phase out Claude or choose other vendors without invoking the label, its appeal remains pending, and the article says courts may now scrutinize procurement labels used against AI vendors.
US District Judge Rita Lin stepped in for Anthropic on March 26, 2026, granting a preliminary injunction that puts the Department of Defense’s “supply chain risk” designation of the company on hold. Her written words pull no punches: the designation is “likely both contrary to law and arbitrary and capricious,” and the sweeping measures “appear designed to punish Anthropic” rather than to serve national security. In her own framing, this is “classic illegal First Amendment retaliation.”
For the first time in this standoff between the government and a frontier lab, a court has hit the brakes. Lin had already shown her hand at a hearing held two days before the ruling, observing that the government appeared to be illegally out to “cripple” and “punish” the company. The order kicks in one week later, which leaves the administration time to appeal. The decision also sets a boundary: phasing out Claude and choosing a different vendor remain lawful moves for the Pentagon — what it can no longer do is lean on the supply-chain-risk label as its justification.
What the Injunction Actually Does: A Reset to February 27
By the order’s terms, the situation goes back to how it stood on February 27 — the injunction “restores the status quo” as it was before the punitive directives were issued. Some quick background: the Pentagon, rebranded the Department of War in the Trump era, had been working with Claude for roughly two years, including on sensitive documents and classified-data analysis, before concluding that the company “could not be trusted” because it insisted on restrictions around use. On February 27, Trump told federal agencies to stop using Anthropic technology, and that same day Defense Secretary Pete Hegseth applied the supply-chain-risk label — a tag normally reserved for foreign adversaries. Per WIRED, the designation brought Claude’s use across the government to a standstill and hurt Anthropic’s sales and reputation.
Retaliation or National Security? Lin’s Step-by-Step Answer
Lin builds her argument in three moves. To begin with, the official rationale does not hold: “the record strongly suggests” the government’s stated reasons were pretextual. Next, if command integrity had truly been the worry, the Pentagon could have simply stopped using Claude and had no need for a blacklisting tag. Finally, she dismissed any suggestion that Anthropic might ideologically sabotage the military: there is “no legitimate basis to infer from Anthropic’s forthright insistence on usage restrictions” that the company would sabotage anything. The record, as she reads it, points the opposite way — Anthropic is being punished for criticizing the government’s contracting position in the press.
How the Dispute Began: The Red Lines Around Claude’s Use
The quarrel traces back to the usage boundaries Anthropic draws for its technology, and those boundaries have never shifted: Claude is off-limits for fully autonomous weapons — where AI makes the final battlefield targeting call with no human in the loop — and off-limits for mass domestic surveillance. On March 9, the company took the fight to federal court with two lawsuits naming multiple officials, Hegseth and the Secretary of State among them, contending that the sanctions are unconstitutional. Once the ruling landed, Anthropic said it welcomed the court’s finding that it is likely to succeed on the merits, and stressed that its focus remains working productively with the government.
What the Ruling Leaves Unsettled
Keep the scope in view: this is a preliminary injunction, not a final verdict, and Lin has set no date for a merits trial. POLITICO reports that the delay in the order taking effect was designed to preserve room for an appeal; follow-up coverage indicates the Pentagon has now filed one, and its CTO continued to describe the ban as in force after the decision. Anthropic’s second case — over a separate statute barring it from supplying software to the military — remains pending before a federal appeals court in Washington, DC. Meanwhile, agencies can still cancel contracts or direct contractors to swap out Claude, provided they do not invoke the designation as the reason.
What the Decision Signals for the AI Industry
Three points worth carrying away. One: procurement weapons like the supply-chain-risk label no longer come free — courts have shown they will probe the motive and legality behind them, which hands vendors a genuine line of defense and gives agencies a reason to pause before reaching for the blacklist. Two: safety red lines now carry judicial weight — a lab’s insistence on usage restrictions cannot be recast as evidence of sabotage, a precedent that matters for every company balancing the government market against its own policies. Three: nothing is settled yet — the appeal, the second lawsuit, and the Pentagon’s transition period are all still in motion. In the months ahead, the rulebook for government AI procurement will be written in courtrooms, not press conferences.