Anthropic, Pentagon clash over AI use. Here’s what each side wants

Anthropic has been at odds with the Department of Defense over how its AI models should be used, and its work with the agency is “under review,” a Pentagon spokesperson told CNBC.

Award given to 5-year-old startup 200 million dollars We signed a contract with the Ministry of Defense last year. As of February, Anthropic became the only AI company to deploy its models into the agency’s secret networks and provide customized models to national security clients.

But Emil Michael, the undersecretary of defense for research and engineering, told a summit in Florida on Tuesday that negotiations on “going forward” terms of use were deadlocked.

Anthropic wants assurances that its models won’t be used for autonomous weapons or “mass spying on Americans.” a report From Axios.

The DoD, by contrast, wants to use Anthropic’s models without restrictions “for all lawful use cases.”

“If any company doesn’t want to comply with this, that’s a problem for us,” Michael said. “This can create a dynamic where we start using them and get used to how these models work, and when we need to use it in an emergency, we are prevented from using it.”

It’s the latest wrinkle in Anthropic’s increasingly strained relationship with the Trump administration, which has been openly critical of the company in recent months.

Venture capitalist David Sacks, who serves as the administration’s AI and crypto czar, accused Anthropic of promoting “woke AI” for its stance on regulation.

An Anthropic spokesman said the company had “productive discussions in good faith” with the Department of Defense about “how to properly address these complex issues.”

“Anthropic is committed to using border AI to support U.S. national security,” the spokesperson said.

The startup’s competitors are OpenAI, Google and xAI were also awarded contract awards of up to $200 million from the Department of Defense last year.

Those companies agreed to allow the Defense Department to use their models for all lawful purposes within the military’s unclassified systems, and one company agreed to “all systems,” according to a senior Defense Department official who asked not to be identified because the negotiations are confidential.

If Anthropic ultimately does not agree to the Department of Defense’s terms of use, the agency could label the company as a “supply chain risk,” which would require vendors and contractors to certify they are not using Anthropic models, the person said.

This name is usually reserved for foreign enemies, so it would be a complicated blow to Anthropic.

The company was founded in 2021 by a group of former OpenAI researchers and executives and is best known for developing a family of AI models called Claude.

Anthropic announced earlier this month that it had closed a $30 billion financing round at a valuation of $380 billion; That’s more than double its value since its last rise in September.

WRISTWATCH: Anthropic releases Sonnet 4.6