News

The Pentagon has moved to classify AI company Anthropic as a supply-chain risk following an order from U.S. Secretary of Defense Pete Hegseth. The decision came after negotiations between Anthropic and the Defense Department stalled over the company’s refusal to permit certain military uses of its Claude model.

Anthropic said the classification follows months of bargaining that reached an impasse when the company sought two carve-outs limiting lawful uses of Claude: prohibition on enabling mass domestic surveillance of Americans and a ban on fully autonomous weapon systems. The startup insisted that neither form of usage should be permitted and that efforts to coerce a change in that policy would not succeed.

The dispute escalated into a broader federal directive and public messaging. President Donald Trump posted on Truth Social that he had instructed all federal agencies to remove Anthropic technology from their operations within six months. Secretary Hegseth subsequently used X to order contractors, suppliers, and partners working with the U.S. military to immediately halt any commercial dealings with Anthropic, and to instruct the Department of War to label the company a national security supply-chain risk.

The designation capped weeks of talks about whether the U.S. military may deploy Anthropic’s AI models. Anthropic reiterated in public statements that while it supports lawful foreign intelligence and counterintelligence missions, it opposes using its systems for sweeping domestic surveillance, arguing such use conflicts with democratic norms and introduces novel risks to civil liberties. The company also raised concerns about the technology’s maturity and reliability for applications like autonomous weapons.

Anthropic criticized what it described as the department’s stance that vendors must permit “any lawful use” of their models and strip away internal safeguards, a demand it said was tied to an effort to create an “AI-first” combat capability. A Pentagon memorandum released last month was reported to criticize diversity and inclusion considerations in model tuning and to urge reliance on models without usage-policy constraints that could limit military application.

In response to the designation, Anthropic called the move legally questionable, warning it could set a harmful precedent for American companies negotiating with government customers. The company noted that under 10 U.S.C. 3252 a supply-chain risk designation can only apply to the contractor’s participation in Department of War contracts and should not bar other customers from using Claude.

Sean Parnell, the Pentagon’s principal spokesperson, denied that the department sought to employ AI for mass domestic surveillance or to deploy weapons systems without human oversight, calling such claims false. Parnell urged Anthropic to permit the Pentagon to use its models for all lawful purposes, saying that the request was intended to avoid putting military operations and service members at risk and that the department would not accept a vendor dictating operational terms.

The standoff has split opinion across the tech sector. Hundreds of employees at Google and OpenAI reportedly signed an open letter urging their companies to support Anthropic in its dispute with the Pentagon. Separately, xAI CEO Elon Musk publicly sided with the administration, asserting strong criticism of Anthropic’s stance.

Meanwhile, OpenAI’s CEO Sam Altman disclosed that OpenAI had reached an agreement with the Department of Defense to allow its models to be used on the DoD’s classified network, and he urged that similar terms be extended to other AI providers. Altman restated that AI safety and broadly shared benefits are central to OpenAI’s mission, and singled out prohibitions on domestic mass surveillance and requirements for human responsibility in the use of force-including for autonomous weapon systems-as key safety principles reflected in the DoD agreement.