Pentagon threatens to blacklist Anthropic over AI weapons limits

Defense Secretary Pete Hegseth met with Anthropic CEO Dario Amodei at the Pentagon on Tuesday, February 24, 2026, and issued a blunt ultimatum: allow Claude to be used for all lawful military purposes or lose the contract by 5:01 PM on Friday, February 27. Hegseth also threatened to designate Anthropic a supply chain risk — a designation that would bar any company with Pentagon contracts from using Anthropic products — or to invoke the Defense Production Act to compel compliance. The meeting was cordial in tone, with no raised voices; Hegseth praised Claude and told Amodei the Pentagon wanted to work with the company.

Anthropic held firm on two specific red lines throughout the negotiation: it would not allow Claude to autonomously select or strike targets in lethal weapons systems, and it would not allow Claude to power mass domestic surveillance of American citizens. These were not new demands — Anthropic had included them in every version of its proposed contract language. The Pentagon sent what it called a 'last and final offer' on Wednesday night; Anthropic rejected it Thursday, with Amodei writing publicly: 'Threats do not change our position: we cannot in good conscience accede to their request.'

The Pentagon awarded contracts worth up to $200 million each to four AI companies — Anthropic, Google, OpenAI, and xAI — in July 2025 to develop classified frontier AI capabilities. Anthropic was the first and only one of the four to have its models cleared for classified military networks, working through partner Palantir Technologies. The other three initially operated only in unclassified systems. The day before the Hegseth-Amodei meeting, xAI agreed to allow its Grok model to be used for 'any lawful use' in classified settings, leaving Anthropic as the sole holdout.

Claude was reportedly used during the January 3, 2026 U.S. special operations mission in Caracas, Venezuela that resulted in the capture of Venezuelan President Nicolas Maduro and 83 deaths, including 47 Venezuelan soldiers. The Wall Street Journal and Axios both reported Claude's involvement based on anonymous sources. Amodei told Hegseth directly that Anthropic had not raised concerns with Palantir about Claude's use in the operation and said it had not been an issue operationally.

Pentagon Undersecretary Emil Michael had led the military's negotiating team for months and stated publicly in mid-February that talks with Anthropic had stalled. His core argument was that once Claude is deployed inside classified systems, the Department of War bears full legal responsibility for outcomes and must have complete operational authority — no private company veto over individual use cases. 'You can't lead tactical ops by exception,' a Pentagon official told NPR. 'Legality is the Pentagon's responsibility as the end user.'

Hegseth and senior Pentagon officials labeled Anthropic's position 'woke AI' — the same term used by Hegseth in his January 9 memo calling for an AI-first warfighting force 'without ideological constraints that limit lawful military applications.' White House AI czar David Sacks had been publicly attacking Anthropic since October 2025, accusing the company of running a regulatory capture strategy. The Pentagon's rebranding as the Department of War under the Trump administration is part of the broader cultural shift Hegseth has championed.

The Defense Production Act threat raised genuinely contested legal questions. The law clearly lets the president compel private companies to prioritize national defense production. What it has never done is compel a company to build a fundamentally different version of its product stripped of features the company chose to include. A defense consultant told Axios that Anthropic could argue it is providing custom-built software already tailored to government use, not a commercially available product for which the DPA was designed.

In the room on the Pentagon side: Hegseth, Deputy Secretary Steve Feinberg, Undersecretary for Research and Engineering Emil Michael, Undersecretary for Acquisition and Sustainment Michael Duffey, chief spokesperson Sean Parnell, and general counsel Earl Matthews. The composition of the delegation — including the Pentagon's top lawyer and its acquisition chief — signaled how seriously the Department of War was treating the confrontation.

The same week as the ultimatum, Anthropic released Version 3.0 of its Responsible Scaling Policy, removing a hard commitment to pause training new frontier models if capabilities outpaced safety controls. The company denied the two events were connected. A safety researcher who resigned February 9, two weeks before the meeting became public, had cited internal pressure to set aside values in favor of commercial and government demands.