A federal judge has blocked a Pentagon initiative that had effectively excluded Anthropic, one of the leading private artificial intelligence labs, from contracts with the U.S. government. The decision came Thursday from U.S. District Judge Rita Lin, who vacated the February 27 designation under which Defense Secretary Pete Hegseth had described Anthropic as a "supply‑chain risk" and made the company ineligible for federal procurements.
In a 59‑page opinion, Judge Lin explained her decision by finding that the measures imposed were "arbitrary, capricious, an abuse of discretion, and otherwise not in accordance with law"; she also concluded that the designation amounted to unconstitutional retaliation. The order additionally struck down an extra measure that barred DoD service providers or vendors from dealing with Anthropic.
The vacatur of the designation also removes other similar sanctions imposed by nine federal agencies, including the Pentagon, the Treasury Department, the State Department and the Department of Homeland Security, according to a reconstruction of the case reported by Wired. Those restrictions had been tied to the same supply‑chain risk assessment.
The dispute began with failed negotiations between the Pentagon and Anthropic over a contract worth roughly $200 million to use Anthropic’s Claude models in military applications. Tensions escalated after reports that versions of Claude were used in the operation that led to the capture of Venezuelan president Nicolás Maduro — an episode that, according to reporting, prompted internal concerns and reports from personnel at third‑party companies such as Palantir.
Anthropic, for its part, had placed usage limits on its models: the company had explicitly opposed the use of its technologies for lethal autonomous weapons and for mass surveillance systems. Hegseth, by contrast, argued that a contractor could not impose constraints on how technology would be used once deployed, stating that the contract would allow "all lawful uses."
When negotiations collapsed, many observers interpreted the supply‑chain risk designation as a punitive move — a reading that now receives support from Judge Lin’s rationale, in which she wrote that, while leaving the Defense Department discretion to choose AI suppliers, the measures imposed on Anthropic were unlawful and unfounded.
Lin added in the opinion that, contrary to the theory of intentional sabotage by the vendor, there is evidence the government was simultaneously in talks with Anthropic about collaborations on new models, citing in particular discussions about a model called Mythos. "None of this is consistent with a genuine fear that Anthropic is a saboteur who would poison its own software to harm national security," the judge wrote, according to press accounts.
The ruling does not require the Pentagon to use Anthropic’s models: the judicial decision removes administrative obstacles created by the designation but leaves intact the Defense Department’s discretion over which technologies and suppliers to select. A Pentagon spokesperson was not immediately available for comment, and the Department is expected to appeal, according to reports.
Anthropic welcomed the ruling. In a statement reported in the press, company spokesperson Danielle Cohen said: "We welcome the court’s decision recognizing the illegality of this supply‑chain designation. We remain focused on working productively with the government to put AI to work for our national security."
Legally, Anthropic filed two suits: one in a California district court, now resolved in the company’s favor, and another in the U.S. Court of Appeals for the D.C. Circuit, which remains pending. The lawsuits challenge the Pentagon’s decision on constitutional grounds as well, alleging violations of the First and Fifth Amendments tied to purported ideological motives and irregular administrative procedures.
Beyond the litigation, the case raises broader, concrete questions about governance of AI technologies: how to balance national security concerns with contractual rights and conditions set by technology providers; what degree of control the government can legitimately require over systems built and updated by private firms; and how far administrative risk assessments can be used to influence market choices.
It also remains unclear what practical effect the ruling will have on the AI model market and federal contracting: although Anthropic regains the ability to compete for contracts, political tensions and the underlying risk assessments could continue to shape agencies’ purchasing decisions. The pending appeal and the docket still open in the Washington court will be the next items to watch to determine whether Judge Lin’s opinion becomes binding precedent or is overturned by higher courts.