March 24, 2026 — A federal judge sharply challenged the Justice Department over the Defense Department’s effort to bar Anthropic from participating in classified systems, calling the government’s actions “troubling” and suggesting they may be punitive.
The dispute centers on Anthropic’s request to prevent the military from using its Claude model for mass surveillance of Americans or in fully autonomous weapons. The Biden administration argues the government must be able to use Claude for “all lawful purposes.” After talks broke down, the Pentagon designated Anthropic a “supply chain risk” and moved to restrict private contractors from employing Claude on military-related work, prompting Anthropic to file suit.
Anthropic contends the supply chain risk label was an unconstitutional punishment for its speech and has asked the court to block both that designation and President Trump’s executive order banning federal agencies from using Anthropic products.
At a Tuesday hearing in San Francisco, U.S. District Judge Rita Lin questioned whether the government’s steps were narrowly tailored to legitimate security concerns. “They don’t really seem to be tailored to the stated national security concern,” she said, adding that if the worry were simply the integrity of military operations the department could stop using Claude itself. “It looks like defendants went further than that because they were trying to punish Anthropic,” she said. “One of the amicus briefs used the term ‘attempted corporate murder.’ I don’t know if it’s murder, but it looks like an attempt to cripple Anthropic.”
Justice Department attorney Eric Hamilton told the court the designation does not prevent companies that contract with the military from using Anthropic’s model for unrelated commercial work. He acknowledged a widely shared post from Defense Secretary Pete Hegseth saying “no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic,” but said the department lacks authority to terminate contractors for having separate commercial relationships with Anthropic and does not plan to do so. Anthropic attorney Michael Mongan said Hegseth’s post nevertheless created “profound uncertainty” and damaged the company.
The statute cited by the government describes a supply chain risk as a threat that an adversary might sabotage, insert unwanted functions into, or otherwise subvert a national security system. Hamilton told the court the Pentagon labeled Anthropic a risk because negotiations and public disputes left officials unable to trust the company and raised concerns about a “risk of future sabotage,” including scenarios in which software could be manipulated or a “kill switch” installed.
Lin pressed whether the government was effectively saying a company could be designated a supply chain risk for being “stubborn” or asking “annoying questions.” Mongan denied Anthropic had the ability to tamper with deployed software or surveil users and argued a real saboteur would sign the government’s contract terms rather than engage in a public fight. He also noted it was inconsistent for the Pentagon to negotiate with Anthropic up until the end if Anthropic posed a grave threat.
Judge Lin said her forthcoming ruling will focus on whether the supply chain risk designation was lawful; she plans to issue a decision in the coming days.
The case highlights broader tensions over how AI can be used by the military and how much control private developers should retain. Anthropic has been the only AI company whose technology was deployed in classified U.S. military systems. CEO Dario Amodei has said he wants to work with the military but insists on two “red lines”: banning mass surveillance of Americans and prohibiting fully autonomous weapons that can strike without human input. He warns that surveillance capabilities are outpacing legal protections and that current reliability for autonomous weapons is inadequate.
The Pentagon maintains it has no interest in using Anthropic’s technology for mass surveillance or fully autonomous weapons and says those applications are already illegal or barred by policy. Pentagon officials argue that decisions about lawful military uses of AI should not be dictated by private companies and have accused Anthropic of attempting to impose its values on government operations. Former Pentagon CTO Emil Michael publicly criticized Amodei’s stance on social media, calling it a “God-complex.” Judge Lin described that public debate as “fascinating public policy” but reiterated that the court’s task is to evaluate the legality of the supply chain risk label.