A senior Pentagon official revealed that U.S. Defense Secretary Pete Hegseth is "imminent" in severing commercial ties with Anthropic and has classified the artificial intelligence company as a "supply chain risk party" - which means that any company wishing to do business with the U.S. military must sever ties with the company.
"It will be very troublesome to cut ties, but we will make them pay for pushing us to this point," the senior official said.
The importance of this is that such punitive measures are usually reserved for foreign hostile states.
"The War Department's partnership with Anthropic is under review," chief Pentagon spokesman Sean Parnell told Axios. "Our nation requires partners willing to help our warfighters win any fight. At the end of the day, this is about the safety of our troops and the American people."
Anthropic's Claude is currently the only artificial intelligence model that can be used in military classified systems, and is also a leading product in many commercial applications around the world. Pentagon officials have praised Claude's abilities.
As Axios reported on Friday, the software has been deeply integrated into the US military system - the US military used Claude during the raid against Maduro in January this year.
core of contradiction
Anthropic has been engaged in intense negotiations with the Pentagon for months over the terms of the military's use of Claude.
Anthropic CEO Dario Amodei attaches great importance to such issues, but he is also a pragmatist.
Anthropic is prepared to relax its existing terms of use, but wants to ensure its tools are not used for mass surveillance of Americans or the development of weapons that can fire without human intervention.
The Pentagon believes that these restrictions are too stringent, saying that there is a lot of ambiguity and that such terms cannot operate normally. Pentagon officials have insisted in negotiations with Anthropic and three other large artificial intelligence labs - OpenAI, Google and xAI - that the military can use its tools for "all lawful purposes."
A source with knowledge of the situation said that senior Defense Department officials have been dissatisfied with Anthropic for a long time and are taking the opportunity to publicly attack.
other side's point of view
Current mass surveillance laws do not take into account the impact of artificial intelligence. The Pentagon already collects vast amounts of information about people, from social media posts to concealed gun permits, but there are concerns that artificial intelligence will greatly enhance this power and make it possible to target civilians.
An Anthropic spokesperson said: "We are engaged in good faith and productive dialogue with the U.S. Department of War on how to continue to cooperate and appropriately address these new and complex issues."
The spokesperson reiterated the company's commitment to using cutting-edge artificial intelligence for national security, noting that Claude is the first model to be used in classified networks.
Another Anthropic official said: "Existing laws prohibit domestic mass surveillance, but these laws are completely unable to keep up with the development of artificial intelligence."
The official gave an example:
"Artificial intelligence can be used to analyze all public information on a large scale. Legally speaking, the U.S. Department of War has the right to collect public information - the so-called 'open source intelligence', including everything on social media, public forums, and online news. This has always been true, but it was previously limited by manpower and limited scale."
"With artificial intelligence, the U.S. Department of War can continuously monitor and analyze the public posts of every American, and cross-reference them with voter registration rolls and assembly permit records to automatically flag civilians who live near military bases, have criticized military policies online, own guns, or participated in rallies."
potential impact
If Anthropic is listed as a supply chain risk party, all large companies that do business with the Pentagon must prove that they do not use Claude in their workflows.
Given Anthropic's widespread reach, some of these companies are almost certainly using it. The company recently said that eight of the top ten U.S. companies are using Claude.
The contracts the Pentagon is threatening to cancel are worth up to $200 million, a tiny fraction of Anthropic's $14 billion in annual revenue.
Key difficulties
A senior government official said competing models were "only slightly inferior" in government-specific scenarios, complicating sudden replacements.
subsequent impact
The Pentagon's tough stance on Anthropic also sets the tone for its negotiations with OpenAI, Google, and xAI. All three companies have agreed to lift the protection restrictions for use by the military in unclassified systems, but not yet for more sensitive classified work.
A senior administration official said the Pentagon was confident the other three companies would accept the "all lawful uses" standard. But a source familiar with the negotiations said many issues remain unresolved.