Anthropic CEO Dario Amodei apologized Thursday for the tone of a leaked internal memo criticizing the Trump administration, as the Pentagon formally designated the AI company a supply-chain risk.

The news escalated a fight over whether the military can use commercial AI tools for "all lawful purposes."

The dispute has become a test of who sets the limits on military use of artificial intelligence, at a moment when Anthropic says Claude, a next-generation AI assistant, is already being used for intelligence analysis, operational planning, and other national security work, Axios reported.

The leaked note, as described by The Information and other outlets, said Anthropic had not given "dictator-style praise to Trump," criticized OpenAI’s Pentagon deal as "safety theater," and argued the dispute centered on Anthropic’s refusal to drop safeguards against domestic surveillance and fully autonomous weapons.

In a blog post, Amodei said the leaked note "does not reflect my careful or considered views" and called it "an out-of-date assessment of the current situation."

He said Anthropic's "most important" priority is "making sure that our war fighters and national security experts are not deprived of the important tools in the middle of war."

A senior Pentagon official said the Department of War had "officially informed Anthropic leadership the company and its products are deemed a supply chain risk, effective immediately."

The official added that "the military will not allow a vendor to insert itself into the chain of command by restricting the lawful use of a critical capability and put our warfighters at risk."

Anthropic still plans to challenge the designation in court, while arguing the action is narrow and affects only certain defense-related uses.

Microsoft reached a similar conclusion, saying Anthropic products can still be offered to customers outside the Department of War through platforms including Microsoft 365, GitHub, and Microsoft AI Foundry.

The fight grew out of a broader clash over AI guardrails.

In a Feb. 26 statement, Amodei said Anthropic had worked to deploy its models across classified government networks and other national security settings, but drew lines at mass domestic surveillance and fully autonomous weapons.

OpenAI said in a March 2 update on its Pentagon agreement that its tools could not be used for domestic surveillance of U.S. persons or to direct autonomous weapons, and that it did not think Anthropic should be labeled a supply-chain risk.

Anthropic said Thursday that the law cited in the dispute, 10 U.S.C. Section 3252, is limited to procurement actions meant to reduce supply chain risk and requires less intrusive measures to be considered first.

It has been reported that the Pentagon was still actively using Claude to support military operations, including in Iran.

The Associated Press contributed to this report.

• Broadcom Rises 7% on $100B in AI Chip Sales in 2027
• OpenAI Looking at Contract With NATO