Judge Rules Pentagon's Anthropic Blacklist Unlawful

Anthropic wore the Pentagon's "supply chain risk" label for roughly five months. On Thursday evening, August 28, Judge Rita Lin of the U.S. District Court for the Northern District of California ordered it removed, ruling that the designation was unlawful retaliation for the company's public criticism of government AI policy and a violation of the First Amendment.

How the fight started

In February, Anthropic CEO Dario Amodei publicly refused the Department of Defense's contract terms. The company insisted on two contractual limits on its Claude models: no use in fully autonomous weapons, and no use for mass domestic surveillance. The Pentagon wanted those limits gone and access for "any lawful use." When talks collapsed, the department classified Anthropic as a supply chain risk in March, a label that effectively shut the company out of federal contracts.

The weight of that label is easy to underestimate. Supply chain risk designations are normally a tool for hardware and software suppliers suspected of ties to foreign governments; applying one to an American AI lab, over the usage limits in its own terms, had no precedent. The move drew a reaction inside the industry at the time, too: a group of Google and OpenAI employees published an open letter in late February backing Anthropic's position.

What the ruling says

The sharpest line in the opinion: "The empty invocation of national security is not a blank check to punish and retaliate against government critics." Judge Lin found no "articulable basis" in the record for believing Anthropic would sabotage its own model. What the evidence did show, she wrote, was a desire to make "a public example out of Anthropic for its 'arrogance' in criticizing the government."

The decision rests on two constitutional grounds and one administrative one. The First Amendment claim is retaliation against protected speech. The Fifth Amendment claim is that Anthropic was denied the "pre-deprivation process" it was owed before losing the right to bid. And the designation itself was, in the court's words, "arbitrary and capricious," the standard formula for striking down an agency action.

Not the last word

A second case, before a federal appellate court in Washington, is still pending and covers different regulatory grounds the department cited. Anthropic has won in San Francisco, but the label is not fully gone until that second file closes. Whether the Pentagon will appeal was not known at the time of writing.

Anthropic's statement was measured: the company said it welcomed the ruling and remains "focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology." The timing matters. The company is preparing an autumn IPO, and as we reported last week, its filing will list even public backlash against AI as a risk factor. A federal blacklist would have been a far heavier line in that section.

The precedent, in one sentence

A government agency cannot punish a vendor for the safety limits it writes into its own contracts, at least not without a reason it can articulate in court. For any company buying AI, the practical reading is that a provider's red lines (Anthropic's are autonomous weapons and mass surveillance) are now contract terms with teeth, ones that even the largest customer on earth could not override by fiat. That cuts both ways: it protects a vendor's principles, and it means you should read those clauses before you sign, because the same logic could one day cover data processing, facial recognition or employee monitoring in your own deployment.

Sources: Forbes, CNN, The Decoder