A federal judge in San Francisco ruled Thursday that the Pentagon's decision to blacklist Anthropic as a national security "supply-chain risk" was unconstitutional retaliation, vacating sanctions imposed by nine federal agencies against the AI lab. [1] U.S. District Judge Rita Lin found that Defense Secretary Pete Hegseth's Feb. 27 designation — which barred defense contractors from using Anthropic's Claude models in any capacity — violated the company's First Amendment rights, calling the measures "illegal and baseless" in a 59-page ruling. [1]

The dispute traces to a use case the Pentagon wanted and Anthropic refused. According to Wired, the rift began after reports that the U.S. had used Claude in the military operation that captured Venezuelan president Nicolás Maduro; a Palantir employee subsequently relayed concerns from an Anthropic staffer about how the models had been used to U.S. officials. [1] Negotiations over a $200 million contract to use Claude for military applications broke down in February after Anthropic insisted on limits barring its models from lethal autonomous weapons and mass surveillance systems, while Hegseth's Pentagon rejected any restrictions, arguing a contractor could not dictate how deployed technology gets used and demanding "all lawful use." [1][2]

Lin's opinion goes out of its way to separate two questions the ruling does not treat as equivalent. She vacated the sanctions against Anthropic and the penalties barring contractors from working with the company — but she explicitly affirmed that the Pentagon remains free to choose a different AI vendor entirely and is not required to reinstate its relationship with Anthropic. [1] That distinction — a due-process win, not a mandate to use Claude — is the part headlines calling this simply an "Anthropic victory" compress away. CNBC's own reporting acknowledges the limits directly: Anthropic's parallel case in the D.C. Circuit, filed to challenge a second, distinct designation, remains unresolved, meaning the company technically remains a supply-chain risk under that separate action until that case concludes. [2]

Lin's reasoning centered on retaliatory motive rather than national-security substance. She wrote that the government had argued Anthropic's "increasingly hostile manner through the press" and its criticism of the Pentagon's AI policy meant it could not be trusted to "ensure the integrity of its models" — and rejected that reasoning outright. "Neither the Constitution nor the federal statute invoked by Defendants allows them to impose sweeping penalties based principally on Anthropic's critique of the Administration's views," she wrote. [2] Lin also pointed to an inconsistency in the government's own conduct as evidence the "supply-chain risk" designation was pretextual: she noted that even now, government officials are discussing collaboration with Anthropic on its new model, Mythos, in sensitive contexts — a fact she said is "not consistent with a genuine fear that Anthropic is a saboteur who would poison its software to harm national security." [1]

Anthropic framed the ruling as narrow and forward-looking rather than a settling of scores. "We welcome the court's ruling that this supply-chain risk designation was unlawful," spokesperson Danielle Cohen told Wired. "We remain focused on working productively with the government to harness AI for our national security." [1] A Pentagon spokesperson could not immediately be reached for comment, and Wired reported the department is expected to appeal. [1]

The ruling lands at a commercially consequential moment for Anthropic. CNBC reported the decision clears a significant hurdle as the company marches toward what's expected to be a near-record IPO, potentially reopening business opportunities that had been cut off by the blacklisting if the Pentagon relationship can be reestablished. [2] That framing — due-process vindication as a business catalyst — is where CNBC's coverage puts its emphasis, distinct from Wired's focus on the free-speech precedent and the Maduro-raid origin story.

Both outlets, though, leave the harder structural question mostly unexamined: what recourse an AI vendor actually has when a government agency wants a capability the vendor won't provide. Lin's ruling establishes that the government cannot punish Anthropic for saying no. It does not establish that the government must accept no for an answer — it can simply take its business, and its lethal-autonomous-weapons use case, to a competitor with fewer objections. The precedent protects a company's right to refuse; it does not touch the underlying leverage a government customer holds when it controls both the contract and, increasingly, the AI oversight framework its own vendors now operate under. [1]

-- THEO KAPLAN, San Francisco