Federal judge vacates the Pentagon's Anthropic blacklist as First Amendment retaliation, calling it "illegal and baseless"

The Pentagon's Anthropic blacklist is dead. The "any lawful use" fight that triggered it is not.

A federal judge has vacated the Pentagon's blacklisting of Anthropic, ruling that the "supply-chain risk" designation was First Amendment retaliation for the AI lab's refusal to drop usage limits from a $200 million military contract 1 2. Judge Rita F. Lin of the Northern District of California stripped improperly imposed sanctions from nine agencies, including the Pentagon, Treasury, State and Homeland Security, and called the measures "arbitrary, capricious" and "illegal and baseless" in a 59-page ruling handed down Thursday 1 2. "The empty invocation of national security is not a blank check to punish and retaliate against government critics," she wrote 2. The Pentagon is expected to appeal 1, capping a designation that spent only 27 of its 181 days at full force.

The 27-day blacklist that took 181 days to kill

Count the days. Defense Secretary Pete Hegseth designated Anthropic a supply-chain risk on February 27 1. Judge Lin blocked the designation's enforcement on March 26 and ordered the administration to rescind it 3. The final vacatur came Thursday 1. That is 181 days with the designation on the books, but only the first 27 of them, about four weeks, at full force. For the remaining five months, the government defended a measure a federal judge had already stopped, while sanctions sat on nine agencies 1 and the Pentagon's AI business went to seven other labs, including Google, Microsoft, OpenAI and SpaceX 2.

The endpoint lands with an irony: President Donald Trump's directive gave federal agencies a six-month phase-out period to stop using Anthropic products 4, and the final ruling arrived 181 days after the designation, at the mark that clock was set to reach.

The two phases of the fight were not equal, and their ratio is the story. It took 27 days for Anthropic to freeze the blacklist: Lin's injunction on March 26 ordered the administration to rescind the designation 3. The government then spent 154 more days defending a measure a judge had already stopped, until the vacatur on Thursday, August 27 1. Divide one calendar by the other: the designation was enforceable for about 15 percent of its 181-day life, and Washington litigated a stopped measure for 5.7 times as long as the measure ever ran.

How six months of escalation played out

• After reports that the US had used Claude in the operation to capture Venezuelan president Nicolás Maduro, a Palantir employee relayed concerns from an Anthropic staffer to US officials about how the models had been used 1.
• Over the winter, Hegseth moved to renegotiate every AI lab's military contract to permit "any lawful use." Most labs signed; Anthropic held two limits: no mass surveillance of Americans and no lethal autonomous weapons, systems able to kill without human oversight 2. Hegseth rejected any curbs, saying a contractor could not dictate how the technology was used once deployed 1.
• Less than 24 hours before the administration's final ultimatum, Anthropic CEO Dario Amodei publicly refused to change the company's stance 2.
• February 27: Hegseth designated Anthropic a supply-chain risk and barred military contractors and suppliers from doing business with the lab 1. Trump attacked the "Leftwing nut jobs at Anthropic" in the post directing the phase-out 4.
• March: Anthropic filed two suits, one in California district court and one at the US Court of Appeals for the District of Columbia, alleging First and Fifth Amendment violations on ideological grounds 1.
• March 26: Judge Lin blocked enforcement, calling the measures an attempt to cripple Anthropic 3.
• Thursday: the 59-page vacatur lifted the sanctions across nine agencies while affirming the Pentagon may keep buying from other vendors 1.

OpenAI accepted Anthropic's red lines the day of the designation

Late on Friday, February 27, the day of the designation itself, OpenAI chief Sam Altman announced a Pentagon agreement covering the department's classified network that included, in his words, prohibitions on "domestic mass surveillance and human responsibility for the use of force," extending to autonomous weapon systems, plus technical safeguards and embedded engineers 4. Altman said he had asked the Department of War to extend the same terms to every AI company 4. More than 60 OpenAI employees and 300 Google employees had signed an open letter backing Anthropic's position 4.

That sequence reframes the standoff: the limits Hegseth framed as Anthropic seizing "veto power over the operational decisions" of the US military 4 were the deal terms OpenAI signed onto that same day. What got punished, on the record Judge Lin assembled, was the insisting: her March order quoted Department of War records showing the designation answered Anthropic's "hostile manner through the press," classic illegal First Amendment retaliation in her words 2.

The same pattern undercut the saboteur theory. The Pentagon justified the designation by arguing an AI lab able to alter or disable its software in wartime would "introduce unacceptable risk" 1. By April, the National Security Agency, whose parent agency is the Defense Department, was reportedly using Mythos Preview, a cybersecurity-focused frontier model Anthropic withheld from public release and limited to roughly 40 organizations, to scan systems for exploitable vulnerabilities 5. Lin folded that into the ruling: a government that genuinely feared Anthropic as a saboteur, she wrote, would not still be discussing collaboration with the lab on Mythos in an array of sensitive contexts 1.

Lin's ruling assembled a third posture that sits oddly beside the other two. Separately, the administration's AI oversight framework placed Anthropic's advanced models under review for their powerful capabilities, an involvement Lin herself cited in the ruling 1. One government held three postures at once: too risky to contract with, formal enough to put under its own review, and open to negotiating for Mythos.

What changed Thursday, and what did not

What changed, for every AI lab negotiating with Washington, is the price of escalation. The supply-chain-risk designation, typically reserved for foreign actors 3, has now been aimed at an American vendor over contract posture and vacated as unconstitutional retaliation 1. An agency reaching for it in a procurement dispute hands the vendor a 59-page exhibit. Labs watched Anthropic hold two red lines, lose the Pentagon's business to seven rivals for six months, and end with the blacklist struck down: that trade now has a number attached.

What did not: the Pentagon is not required to use Anthropic's models and can keep buying from the seven labs it signed 1; most AI labs had already accepted the "any lawful use" terms 2; the appeal is expected 1; and Anthropic's second suit, at the DC appeals court, is ongoing 1. What those lines are worth in the next contract is still being litigated: one case ongoing, one appeal expected.