Anthropic’s months-long fight with the Trump administration just produced its most consequential ruling yet — and it’s a split decision that leaves the AI company in an unusually messy legal position, blacklisted in one part of the government while cleared to keep working with the rest of it.
How the Standoff Started
The dispute traces back to February 2026, when Defense Secretary Pete Hegseth pushed Anthropic to let the Pentagon use Claude for “all lawful purposes” on its systems. Anthropic refused, citing its own policy red lines against using its models for fully autonomous lethal weapons or mass domestic surveillance of US citizens. The Pentagon’s response was severe: in March, it designated Anthropic a “supply chain risk” under two separate legal authorities, canceling the company’s existing defense contracts and barring other Pentagon contractors from using Claude in any of their own work with the department.
Anthropic pushed back in court — twice, in two different venues, because the two designations fell under different laws. One case went to federal district court in San Francisco; the other went straight to the D.C. Circuit Court of Appeals. The company argued the government’s action was arbitrary, unconstitutional, and effectively retaliation for refusing to drop its safety guardrails.
Two Courts, Two Different Answers
For a while, it looked like Anthropic might be winning. In August, U.S. District Judge Rita Lin in San Francisco ruled that one of the two designations — the broader, government-wide ban that would have barred essentially all federal agencies and contractors from touching Anthropic’s technology — was unlawful, finding it amounted to First Amendment retaliation. That ruling stuck: the Trump administration didn’t appeal it, and other federal agencies remain free to work with Anthropic today.
Friday changed the picture. In a 2-1 decision, the D.C. Circuit upheld the second, narrower designation — the Pentagon-specific “supply chain risk” label grounded in public procurement rules. The majority found it was reasonable for the Defense Department to treat Anthropic’s refusal to remove its restrictions as grounds for exclusion. Practically, that means Claude stays barred specifically within the Pentagon, even though the rest of the federal government can still use it. Anthropic called the outcome disappointing and pointed out that a separate court had already found the government’s parallel action unlawful; the company says it’s weighing further options, including asking the full D.C. Circuit to rehear the case en banc, or petitioning the Supreme Court. The Pentagon, for its part, treated the ruling as vindication — Undersecretary Emil Michael celebrated it publicly as a rebuke of Anthropic’s position.
Anthropic has said the blacklisting has already cost it billions of dollars in lost business and damaged its standing at a particularly bad moment: the company is widely expected to pursue an IPO, and a live legal cloud over its relationship with the US government’s largest single customer is not the kind of headline any pre-IPO company wants.
The Bigger Pattern Here
Strip away the legal technicalities, and what’s left is a genuinely unusual confrontation: a major US AI lab publicly refusing a request from its own government’s defense establishment, then getting formally penalized for it — the first time this kind of supply-chain-risk authority has been used against a domestic American company rather than a foreign one. The Trump administration has made no secret of its frustration with Anthropic specifically; the White House has repeatedly dismissed CEO Dario Amodei’s public warnings about AI risk, with Trump himself once calling them a hoax.
That friction points to a genuine and unresolved dividing line running through the AI industry: model providers that build in restrictions around lethal autonomy and mass surveillance are discovering that those very restrictions can become a liability when a government customer wants the technology to work without limits. As militaries around the world race to integrate frontier AI, this case will likely be studied closely by every other AI company weighing how much leverage — and how much risk — comes with holding a firm line against a customer that happens to also be the state.