The clash between Anthropic and the Pentagon seems to be escalating day by day. According to a recent filing, Anthropic has told a federal appeals court that it cannot control its AI models after they are deployed to the Pentagon. The filing challenges a Pentagon decision that labeled the company a supply chain risk. That dispute now shapes Anthropic’s access to new Defense Department contracts.
Court filing centers on control after deployment
Anthropic said it has no visibility into models after Pentagon deployment. It also said no technical “kill switch” exists. The company said the Pentagon can test models before deployment. It argued that testing gives officials a direct review path.
Anthropic’s filing went to a federal appeals court in Washington. The company used that filing to challenge the Pentagon’s designation. The Pentagon labeled Anthropic a supply chain risk in court. It said the company inserts itself into sensitive military use decisions.
Anthropic’s usage rules bar Claude from autonomous weapons deployments. They also bar mass surveillance under the company’s published restrictions. Pentagon lawyers dismissed those limits as red herrings in court. That disagreement helped drive the current legal fight.
Split rulings leave the Pentagon ban in place
A court in Washington denied Anthropic’s request for a pause. However, a California judge granted relief in a parallel case. That split ruling left Anthropic blocked from new Pentagon contracts. Still, the company can work with other agencies.
The Pentagon continues defending its designation in court. The dispute now runs alongside broader federal AI deployment efforts. The Trump administration is moving to deploy its Mythos model widely. That push reaches across the federal government.
At the same time, agency heads are assessing cyber defense needs. They are also weighing how Mythos can protect government systems. Those internal efforts could complicate the Pentagon’s current argument. The department says Anthropic poses a national security risk.
Anthropic’s filing states it cannot direct deployed systems remotely. It also states that the Pentagon controls use after deployment. The next court hearing in the case is scheduled for May 19. That date now marks the next major step in the dispute.
Inside the Anthropic standoff: NSA’s Mythos use and White House talks raise the stakes
The latest report that Anthropic cannot control its AI after Pentagon deployment has pushed its Washington dispute back into view. That filing arrived after reports of a White House meeting and after claims that the National Security Agency uses Mythos Preview. Together, those developments have turned a contract dispute into a wider debate over military access, cyber tools, and government oversight.
Last week, attention first shifted to the White House. Anthropic chief executive Dario Amodei met White House chief of staff Susie Wiles as pressure built around the company’s fight with the Pentagon. That meeting focused on the limits Anthropic wants to place on military use of its systems and the government’s interest in stronger cyber models.
Soon after, another report placed Mythos at the center of the story. As we reported, the National Security Agency is reportedly using Mythos Preview, a restricted cyber model that Anthropic has not released publicly. The report also said Mythos has broader use inside the Department of War, though officials did not describe its exact role.
That claim matters because Anthropic has kept Mythos tightly restricted. The company said the model can find, analyze, and explore software vulnerabilities at scale. Anthropic also said internal testing showed Mythos identified thousands of critical flaws, including zero-day vulnerabilities. Now, a new court filing has added another layer. Anthropic has argued that the Pentagon has a full opportunity to test systems before deployment, which leaves post-deployment use in the government’s hands.