The "supply-chain risk" label, confirmed in a statement by Anthropic, is effective immediately and bars government contractors from using Anthropic's technology in their work for the US military.
But companies can still use Anthropic's Claude in other projects unrelated to the Pentagon, CEO Dario Amodei wrote in the statement on Thursday.
He said the designation has "a narrow scope" and that the restrictions only apply to the usage of Anthropic AI in Pentagon contracts.
"It plainly applies only to the use of Claude by customers as a direct part of contracts with the Department of War, not all use of Claude by customers who have such contracts."
The risk designation follows a months-long dispute over the company's insistence on safeguards that the Defense Department, which the Trump administration calls the Department of War, said went too far.
In his statement, Amodei reiterated the company would challenge the designation in court.
In recent days, Anthropic and the Pentagon have discussed possible plans for the Pentagon to stop using Claude, Amodei said in the Thursday statement. The two sides have talked about how Anthropic might still work with the military without dismantling its safeguards, he added.
However, in a post on X late on Thursday, Pentagon chief technology officer Emil Michael said there was no active Department of Defense negotiation with Anthropic.
Amodei also apologised for an internal memo published on Wednesday by the tech news site The Information. In the memo, originally written last Friday, Amodei said Pentagon officials didn't like the company in part because "we haven't given dictator-style praise to Trump".
The internal memo's publication came as Anthropic's investors were racing to contain the damage caused by the company's fallout with the Pentagon.
The Defense Department did not immediately return requests for comment.
The action represented an extraordinary rebuke by the United States against an American tech company that was earlier than its rivals to work with the Pentagon.
The action comes as the department continues to rely on Anthropic's technology to provide support for military operations, including in Iran, according to a person familiar with the matter.
Claude likely is being used to analyse intelligence and assist with operational planning.
The "supply-chain risk" label now gives Anthropic a status that Washington until now had typically used for foreign adversaries.
Similar US action was taken to remove Chinese tech giant Huawei from the Pentagon's supply chains.
