A Judge Just Told the Pentagon Its Ban on AI Company Anthropic Was Illegal
A federal judge ruled on August 27, 2026, that the Pentagon broke the law when it labeled AI company Anthropic a security risk and tried to ban government use of its products. The 59-page ruling by Judge Rita Lin, issued in San Francisco, blocks certain federal agencies from enforcing President Trump's order to stop using Claude, Anthropic's AI assistant, and other company tools. Politico
Judge Lin, appointed by President Biden, wrote that "the empty invocation of national security is not a blank check to punish and retaliate against government critics." The court found that Pentagon Chief Pete Hegseth acted illegally when he designated Anthropic a supply chain risk. This was the first time a US company had received that label, which the government normally uses on foreign technology companies that might pose espionage or sabotage threats. The government justified its actions largely with a four-page memo. Washington Post
The conflict started in late February, when the Defense Department and Secretary Hegseth pressured Anthropic to remove certain safety features from its AI systems. Anthropic CEO Dario Amodei refused, saying the company would not allow its AI to be used for mass surveillance or the development of weapons that operate without human control. Days later, President Trump ordered all US government agencies to stop using Claude and other Anthropic services within six months. Engadget
Anthropic sued the US government in two federal courts, challenging two separate actions. The lawsuit against the Pentagon was filed on March 9. On March 26, Judge Lin granted a temporary court order blocking the ban while the case continued. Reuters In June, the Trump administration denied that it had retaliated unlawfully against the company. Reuters
This week's ruling gave Anthropic a win on one of its two legal challenges. The company still carries the supply chain risk label on a second, separate track, which has not yet been decided.
The case has been widely seen as the first major clash between the US government and an AI company. The core question was whether national security powers, designed to screen foreign technology companies for spying or sabotage risks, could be redirected to pressure an American software company over decisions about its own product safety.
While the legal battle played out, Anthropic's business barely slowed. Second-quarter revenue jumped to $11.5 billion, about fourteen times what it earned in the same quarter a year earlier. Even if the government's purchasing restrictions had stayed in place, federal contracts would have been a small fraction of that total.
The ruling raises a question that will keep coming up as AI tools become more deeply embedded in government work: where does legitimate national security oversight end and punitive use of government buying power begin? Judge Lin's opinion draws that line clearly. The fact that the government's case rested on a four-page memo, answered by a 59-page judicial decision, suggests the court found the evidence far too thin to justify the action.
The second legal challenge remains unresolved, leaving the situation in partial limbo. Anthropic still holds the supply chain risk label on one track, and the government could appeal the ruling on the other. What is settled, for now, is that a federal court has put a firm limit on the executive branch's power to use supply chain rules against a domestic AI company that refused to weaken its safety features when the Pentagon asked.
The precedent extends beyond Anthropic. Other companies building advanced AI now have a court decision they can point to: refusing government pressure to remove safety features can be defended in court, and the business impact of saying no may be small compared to what private customers are spending. Whether that same logic holds for smaller companies without Anthropic's massive revenue is a different question, and one the industry will likely confront as AI tools spread further into military and classified settings.
