Anthropic built its entire brand around being the safety-first AI company. Now that identity is the reason a federal court just handed it one of the most consequential legal losses in its history.
A federal appeals court in Washington, D.C., ruled 2-to-1 that the Department of Defense acted lawfully when it designated Anthropic a national security supply chain risk. The decision backs Defense Secretary Pete Hegseth and the Trump administration, and it leaves Anthropic cut off from military and defense contractor business at exactly the moment the company was gearing up for a highly anticipated IPO. The company says the ban has already cost it billions in lost contracts.
The roots of this go back to a $200 million deal signed in July 2025, designed to bring Claude models to the Pentagon’s GenAI.mil platform. The contract fell apart when Anthropic refused to drop its internal usage restrictions. Specifically, the company prohibited using Claude for fully autonomous lethal weapons and mass domestic surveillance of American citizens. It also wanted to retain the right to remotely shut down access if it detected use outside those boundaries. The Pentagon said no. When Anthropic held firm, Hegseth formally labeled it a supply chain risk under the Supply Chain Security Act, blocking both military agencies and their contractors from using Claude.
The majority opinion from Trump-appointed Judge Gregory Katsas, joined by Judge Neomi Rao, sided with the military’s core argument. Katsas wrote that corporate models with built-in ethical constraints could shut down unpredictably during active operations or be manipulated in ways that undermine mission integrity. The ruling put decision-making authority squarely with the President and Defense Secretary, not an AI company’s internal policy team.
But Judge Karen LeCraft Henderson dissented sharply. Her argument was direct: enforcing upfront ethical standards is not a supply chain security threat. That’s not a fringe position. It’s also the logic that a separate federal court in San Francisco already accepted. Last month, U.S. District Judge Rita Lin struck down a broader government-wide ban on Anthropic, finding officials had unlawfully retaliated against the company because of its public stance on AI safety. Two courts, two outcomes, one company caught in the middle.
For the broader AI industry, this case is worth watching closely. Anthropic is not alone in building safety constraints into its models. OpenAI, Google DeepMind, and others all maintain some version of acceptable use policies that would likely conflict with unrestricted military deployment. If the Pentagon’s blacklisting of Anthropic holds up through all appeals, it sends a clear signal to any AI company that wants government contracts: your internal guardrails are a liability, not a selling point. That’s a direct tension with the direction most frontier AI labs have publicly committed to.
The D.C. panel has delayed enforcement of the ruling to give Anthropic time to either petition for an en banc rehearing by the full appeals court or take the case to the Supreme Court. Anthropic said it respectfully disagrees with the decision and is evaluating its options. Hegseth, for his part, celebrated on X.
So the fight isn’t over. But Anthropic is now fighting on two legal fronts simultaneously, spending resources in court rather than on customers, and watching a potential IPO hang in the background. The company bet that safety would be a competitive advantage. In the government market, at least, that bet is being stress-tested in ways nobody anticipated.
