A person familiar with the situation told CBS News that trust is eroding between the Pentagon and Anthropic over the use of AI models.
In a meeting at the Pentagon on Tuesday morning, Defense Secretary Pete Hegseth told Anthropic CEO Dario Amodei to provide the military with a signed document by the end of this week granting the company full access to the company’s artificial intelligence models, sources said.
Officials say they are considering invoking the Defense Production Act to force Anthropic to comply with military demands. Axios previously reported on some of what happened at the meeting.
Defense officials want to have complete control over Anthropic’s AI technology for use in military operations, sources told CBS News. The company won a $200 million contract from the Department of Defense in July to develop AI capabilities to advance U.S. national security.
Antropic has repeatedly asked the Pentagon to agree to guardrails that would limit the AI model, called Claude, from conducting mass surveillance of Americans, the people said. Defense officials pointed out that this was illegal and said the military was only seeking licenses to use AI for lawful activities.
Free Press: Are we on an AI cliff?
Mr. Amodei also wants to ensure that Mr. Claude is not used by the Pentagon to determine the ultimate goal of military operations without human involvement, one person familiar with the talks said. Claude is not immune to hallucinations, nor is he reliable enough to avoid fatal mistakes, such as unintentional escalations or mission failures without human judgment, officials said.
But when asked for comment, a senior Pentagon official said: “This has nothing to do with mass surveillance or the use of autonomous weapons. The Pentagon just issued a lawful order.”
The person said Grok, owned by Elon Musk’s xAI, has agreed to use it in sensitive environments, and other AI companies are making similar moves.
At Tuesday’s meeting, Hegseth told Amodei that when the government buys Boeing planes, aerospace companies have no say in how the Pentagon uses the planes. He argued that the same should apply to the military’s use of Claude.
After Mr. Amodei left, officials debated whether to use the Defense Production Act, which allows the government to control domestic industry, in this situation.
But officials say they’re not sure the government can trust Anthropic at this point, and the Pentagon may decide to formally designate Anthropic as a “supply chain risk” in order to force it out of government service, two people familiar with the matter said. Anthropic was the first technology company authorized to work on sensitive military networks.
An Anthropic spokesperson said in a statement: “We continued to discuss usage policies in good faith so that Anthropic can continue to support the government’s national security mission consistent with what our model can do reliably and responsibly.”
Hegseth gave Anthropic a deadline of 5 p.m. Friday.
