Anthropic Denies Possessing ‘Kill Switch’ Over AI Deployed to US Military

(SeaPRwire) –   The company has pushed back against the Pentagon’s claims that it retains any ability to control the Claude AI system once it’s deployed on military networks

Per a recent court submission, AI firm Anthropic maintains it has no backdoor or “kill switch” for its Claude AI after the system is rolled out on the Pentagon’s classified military networks.

Earlier this year, the U.S. military and the tech company became caught up in a policy disagreement: the Pentagon wanted to use the system for “all lawful military purposes,” but Anthropic stuck to its AI safety protocols regarding mass surveillance and the use of fully autonomous weapons.

Eventually, the Pentagon cut ties with Anthropic, labeling the tech firm a “supply chain risk” — a seldom-used designation usually applied to groups connected to Washington’s foreign rivals. This label prevents Anthropic from not just working directly with the U.S. government, but also stops other contractors from using its products.

In a fresh filing with the federal appeals court in Washington, D.C., Anthropic challenged a core claim by the U.S. administration: that the company still had some level of control over Claude AI once it was deployed on classified systems, effectively giving itself an “operational veto.” The firm stated it has “no back door or remote kill switch,” and that its “staff can’t log into a department system to alter or shut down a running model.”

The AI system provided to the Pentagon is a “static” model, Anthropic contended. Once deployed, it “doesn’t degrade or change automatically, and Anthropic can’t send unannounced or unauthorized updates to the model after the department has rolled it out.”

Anthropic was officially labeled a “supply-chain risk to national security” on February 27, around the same time U.S. President Donald Trump claimed it was run by “leftwing nut jobs.” The company has contested this designation in court, and so far the legal fight has produced mixed outcomes.

Earlier this month, the D.C. court turned down Anthropic’s request to put the supply chain risk designation on hold. But in a separate case in California, a court ruled in the company’s favor, temporarily halting the administration’s decision. Due to this split ruling, Anthropic is still prohibited from working with the Pentagon but can keep its partnerships with other agencies as the legal fight continues.