Hegseth Gives Anthropic Friday Deadline to Drop Claude Safeguards for Military Use

Pete Hegseth reportedly gave Anthropic CEO Dario Amodei until Friday evening, February 27, to let the military use Claude without Anthropic’s usual safeguards or face penalties, including the risk of losing Pentagon business. The meeting was described as tense, with one account saying it was intense with back-and-forth arguments and another saying it stayed cordial with no raised voices.

What makes this bigger is no one can solve quickly. Anthropic is not just another vendor. Claude has been the key model in classified defense use, and while the Pentagon is moving faster on alternatives, that does not instantly fix the replacement problem. xAI/Grok and Google is seen as a possible next option, but Claude is still ahead in some defense-related uses.

Anthropic’s position is easier to misread. This is not a company refusing defense work. It already has a Pentagon agreement worth up to $200 million and has publicly framed that work as responsible national security support. Focusing on fully autonomous military targeting and domestic surveillance of Americans.

The timing matters because the clash only grew after Claude was reportedly used via Palantir in the Maduro raid and that gave Pentagon what they need.

The Pentagon’s options sound severe because the pressure campaign has included talk of  labeling Anthropic a supply chain risk that would not just hit their deal with the Pentagon. It could force other defense contractors and partners to certify Claude is not inside their military workflows, turning one dispute into a wider cutoff. And the Defense Production Act talk is even more aggressive because it is a way to force Anthropic to adapt its model to Pentagon needs.

Just before the meeting, Anthropic published a report accusing Chinese labs, including DeepSeek of large-scale distillation attacks to clone Claude's features. Even if that was not aimed at this Pentagon fight, it reinforces Anthropic’s argument that frontier model controls are part of national security too. Reuters and AP reporting also make clear this is no longer just a vendor dispute, but a fight over unrestricted military use, surveillance risk, and who sets the rules once AI is already inside military systems.