The contractor’s veto

Editor's note: The following text is a transcript of a podcast story. To listen to the story, click on the arrow beneath the headline above.

NICK EICHER, HOST: Up next: The Pentagon fires its AI assistant.

Later this morning, artificial intelligence company Anthropic heads to court in northern California for its first hearing in a case against the Pentagon.

MARY REICHARD, HOST: This comes after a contract dispute led the Pentagon to cut ties and blacklist Anthropic as a supply chain risk.

WORLD's Harrison Watters has the story.

HARRISON WATTERS: Anthropic’s lawsuit says the company was founded on the belief that AI development should maximize positive outcomes for humanity. Its founders had left ChatGPT’s developer, OpenAI, over concerns that company wasn’t developing AI responsibly. 

Here’s CEO Dario Amodei on a tech podcast in 2024:

AMODEI: Civilization is going down this path to very powerful AI: what's the way to do it that is cautious, straightforward, honest…

Last June, Anthropic announced it was creating custom models for national security customers. And it was the first to get security clearance to work with the Pentagon.

As the Department of Defense was updating its name to the Department of War last fall, it was also stepping up its use of AI tools.

Defense Under Secretary Emil Michael spoke to CBS earlier this month.

MICHAEL: So we're still at the early stages of using it for Military purposes, but it's helping make every part of the Department of War that it's used in just more efficient.

Then the partnership hit a snag. Here’s Daniel Cochrane, senior research associate at the Heritage Foundation.

COCHRANE: So it appears that in late 2025, the Pentagon was reviewing various contracts with their vendors, and they found that Anthropic imposed these two red lines.

Anthropic was part of a contract with Palantir, a data analysis company that serves the U.S. Military. In the contract, Anthropic said their models could not be used for developing autonomous weapons that could kill without humans pulling the trigger. They also forbade the Department of War from using Claude models to build surveillance systems for monitoring American citizens.

After the U.S. Military seized Venezuelan president Nicholas Maduro in January, Anthropic asked the Pentagon if its models were used during the raid.

Last week, Under Secretary Michael told CNBC that was classified information, and breached the chain of command.

MICHAEL: And implying that if they didn't like the way it was used in that raid, that we might have a problem and have violated the terms of the contract, which meant that they could shut us off at any time.

The Department of War wanted to update the contract. But Anthropic wanted its restrictions to remain intact. 

Here’s CEO Amodei on CBS News:

AMODEI: You know, at one point they sent us language that, you know, appeared on the surface to meet our terms, but it had all kinds of language like "if the Pentagon deems it appropriate" or "to do anything in line with laws." So it didn't actually concede in any meaningful way.

Cochrane says the conflict comes down to where the buck stops.

COCHRANE: Who gets to make that ultimate call? Is that the Defense Department that is ultimately accountable to the President of the United States, who is an elected leader, or is it a private AI company?

At the end of February, the Pentagon gave Anthropic its own red line: remove the restrictions on Claude, or be designated a supply chain risk. That designation would prevent Defense agencies or contractors they do business with from using Anthropic’s services on government projects.

Cochrane says that’s typically reserved for foreign companies.

COCHRANE: So, for example, the Pentagon might say, if you're a contractor that does business with them, you can't use technology from Huawei, because it's a Chinese company, it's controlled by a foreign adversary.

Under Secretary Michael told CNBC the designation was necessary because the Pentagon believes Anthropic’s worldview is not aligned with the U.S. Constitution.

MICHAEL: We can't have a company that has a different policy preference that is baked into the model through its constitution, its soul, its policy preferences, pollute the supply chain…

President Trump later called on all government agencies to stop using Anthropic’s products. But removing Claude from the American government’s systems won’t be as simple as cancelling the contract and hiring a different AI company. 

Here’s Ethics and Public Policy Center fellow Clare Morell:

MORELL: The switching costs are a lot higher, because…you've already invested a lot in training personnel, developing systems, all based on this certain AI model, and they're all slightly different. And so that can then take a lot of time to unwind.

Six months to unwind to be exact, according to the Department of War.

While the Department works with Anthropic to extricate Claude, it’s also stepping up its security clearance partnership with other developers, including Amodei’s former employer, OpenAI.

For its part, Anthropic stands to lose a lot more than its $200 million Pentagon contract, and it’s hoping courts agree. 

Daniel Cochrane at the Heritage Foundation says the Department of Defense has full discretion to hire and fire contractors. But branding a leading American AI developer a supply chain risk could hurt the larger industry.

COCHRANE: No one else can do business with Anthropic if you touch any Pentagon contract, I think that is perhaps too far, but we'll see what the courts say in the end.

The court today will hear Anthropic’s request to put the government’s order on hold while the case proceeds.

Reporting for WORLD, I’m Harrison Watters.

WORLD Radio transcripts are created on a rush deadline. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of WORLD Radio programming is the audio record.