Can Anthropic Make Peace With the Pentagon?

“The artificial intelligence start-up has reportedly resumed talks with the Defense Department over use of its tools, with high stakes for both.

 

Andrew here. The soap opera between the Pentagon, Anthropic and OpenAI continues. And an emerging question is whether the fight is less about A.I. safety and more about politics.

 

Dario Amodei, Anthropic’s C.E.O., said the quiet part out loud in a leaked internal memo, suggesting the company was being targeted by the Pentagon because “we haven’t donated to Trump” and that “we haven’t given dictator-style praise to Trump (while Sam has).”

 

That OpenAI has sought to add stricter terms to its own contract with the Pentagon, a move that also appears to undermine the government’s argument about why it refused a deal with Anthropic. More below.

 

Back to the negotiating table?

 

Less than a week after Defense Secretary Pete Hegseth threatened to essentially blacklist the artificial intelligence company Anthropic, the Pentagon has resumed talks about using its models, The Financial Times and Bloomberg report.

 

Any renewed talks carry high stakes for Anthropic. Its business faces a dire threat if Hegseth’s original threat comes to pass. Stakes are high for the Pentagon, too.

 

It’s unclear how the two sides can compromise. The Pentagon has stressed that it believes it should be allowed to employ A.I. tools for “any lawful use,” though it could agree to some safeguards on mass surveillance and autonomous weapons. It points to a deal it reached with OpenAI — one that Sam Altman, OpenAI’s C.E.O., said addressed concerns like Anthropic’s.

 

But in a blistering memo sent to Anthropic employees on Friday, first reported by The Information, Dario Amodei, Anthropic’s C.E.O., argued that the OpenAI contract appeared to have left the Pentagon plenty of loopholes. OpenAI and the Pentagon’s descriptions of their deal sounded like “80 percent safety theater,” he wrote, adding that it basically gave the department what it wanted.

 

Altman acknowledged flaws in OpenAI’s original contract this week, and announced that his company had negotiated what he called stronger safeguards.

 

Will cooler heads prevail? After Hegseth cut off talks with Anthropic last week, Emil Michael, the Pentagon official who led the negotiations, called Amodei a “liar” with a “God complex.”

 

And Amodei may have to explain strong language in that internal memo, notably his assertions that the Pentagon’s animus was because “we haven’t donated to Trump” and that “we haven’t given dictator-style praise to Trump (while Sam has).”

 

Anthropic has a lot on the line. The company’s annual revenue run rate has reportedly hit $20 billion, more than doubling where it was in late December, after a surge in business adoption of its Claude models.

 

But the prospect of being barred from any business with government contractors — which Anthropic has said it will fight in court — could decimate those gains. (Several defense contractors are already dropping Claude.)

 

It’s a big deal for the Pentagon, too. Until recently, Claude was the only A.I. tool allowed into its classified systems, and it was reportedly used in the military operations in Venezuela and Iran.

 

 Government officials have said Claude is better than some alternatives, and have acknowledged that removing Anthropic tools from their operations will be a messy task.” [1]

 

1. Can Anthropic Make Peace With the Pentagon?: DealBook Newsletter. Andrew Ross Sorkin; Warner, Bernhard; Kessler, Sarah; Michael J. de la Merced; Gallogly, Niko; et al.  New York Times (Online) New York Times Company. Mar 5, 2026.