Azernews.Az

Akbar Novruz

Read more

The artificial intelligence startup Anthropic on Thursday rejected the Pentagon’s demand for unfettered access to its Claude AI model, suggesting it is willing to risk the serious penalties threatened by Defense Secretary Pete Hegseth, AzerNEWS reports via Politico.

“[T]hese threats do not change our position,” Anthropic CEO Dario Amodei wrote in a blog post published on Thursday. He said the company “cannot in good conscience accede to [the Pentagon’s] request.”

Anthropic’s defiance follows unprecedented pressure from the Pentagon to abandon its restrictions on the military’s use of Claude. In a Tuesday meeting with Hegseth, Amodei reiterated his red lines, a ban on the technology’s use to surveil American citizens or to empower autonomous weapons.

Hegseth had threatened to designate Anthropic a risk to the Pentagon’s supply chain if it failed to comply by 5:01 pm Friday. The label is almost always reserved for foreign firms with ties to U.S. adversaries, and could be used by the government to blacklist Anthropic and prevent it from working with other companies.

The Pentagon was simultaneously planning to use the Cold War-era Defense Production Act to compel Anthropic to lift restrictions on the use of its model. The law allows the government to take control of critical private-sector assets in times of war or national emergency.

In a statement, an Anthropic spokesperson, granted anonymity at the company’s insistence, said they received contract language from the Pentagon overnight that “made virtually no progress on preventing Claude’s use for mass surveillance of Americans or in fully autonomous weapons.” The spokesperson said that despite being framed as a compromise, the new language “was paired with legalese that would allow those safeguards to be disregarded at will.”

On Thursday, CBS News reported that the contract language sent overnight to Anthropic represented the Pentagon’s “final offer.” But in his blog post, Amodei suggested a willingness to continue negotiations with the DOD. “Our strong preference is to continue to serve the Department and our warfighters — with our two requested safeguards in place,” the Anthropic CEO wrote.

The company spokesperson said Anthropic “remain[s] ready to continue talks.” It is still possible that a deal will materialize before Friday’s deadline.

It is not immediately clear whether or how the Pentagon will follow through on its threats. A DOD spokesperson did not immediately respond to a request for comment.

Hegseth’s move to invoke the DPA suggests that the Pentagon sees Anthropic’s AI models as critically important to U.S. national defense — a stance some lawyers and AI policymakers said was contradictory, given the Pentagon’s concurrent claim that the company may be a national security risk. The discrepancy was highlighted by Amodei in his blog post.

“These latter two threats are inherently contradictory,” Amodei wrote. “[O]ne labels us a security risk; the other labels Claude as essential to national security.”