JUANA SUMMERS, HOST:
Yesterday, a federal judge handed down a big win for the AI company Anthropic. It ruled the Pentagon could not designate Anthropic a supply chain risk, a label that would have cost it business beyond the Department of Defense. NPR's Geoff Brumfiel has been looking at the ruling, and he's on the line now. Hey.
GEOFF BRUMFIEL, BYLINE: Hi, Juana.
SUMMERS: Geoff, tell us what the judge said about the case.
BRUMFIEL: Yeah, District Judge Rita Lin was absolutely scathing in her rebuke of the Pentagon. She called the evidence that Anthropic could be a national security risk, quote, "slim." And she said the true reason for this designation appeared to be to get back at Anthropic for criticizing the government. She said the Pentagon was seeking to make an example out of the company. Given all that, she wrote, the Pentagon's actions, quote, "constitute unlawful retaliation in violation of the First Amendment." And she said the Pentagon could not use the supply chain risk designation for the company.
SUMMERS: OK, I mean, it sounds like this was a huge win for Anthropic, but I want to take a step back here. Remind me, how did this all begin?
BRUMFIEL: Well, Anthropic, as I'm sure most listeners know, is one of these big AI companies, and I should also mention at this point, they're a financial supporter of NPR. Anyway, their chatbot called Claude was actually the first chatbot cleared to do classified work by the government. Since 2024, according to court records, the intelligence community was using a special version of Claude called Claude Gov.
Now, just like you and me, the government has to sign these terms and conditions, you know, that come up. And in that document was language about how it couldn't use Claude for either mass surveillance of Americans or lethal fully autonomous warfare. So no murder bots or anything like that.
SUMMERS: OK.
BRUMFIEL: The government actually agreed to those terms, but then last fall, the newly renamed Department of War entered into new negotiations and demanded the language be removed. Long story short, Anthropic refused, and Defense Secretary Pete Hegseth designated them a supply chain risk, or at least he tried to.
SUMMERS: I do have to ask, what does it actually mean to be a supply chain risk?
BRUMFIEL: It's a huge deal. Basically, it would mean that not only could the government not use Anthropic, but any company doing business with the government would also have to avoid using Anthropic. The government's argument hinged on the idea that anthropic had some sort of backdoor into its software, and they could shut it down if they didn't like how it was used. Aalok Mehta is with the Center for Strategic and International Studies, a D.C. think tank. He says the Pentagon never really made the case.
AALOK MEHTA: There was very little evidence that this was something Anthropic was intending to do. I think there's very little evidence that it's even technically possible.
BRUMFIEL: And without proof Anthropic was seeking to undermine military operations, the judge ruled this designation was retaliatory.
SUMMERS: Geoff, I just want to get back to what's really at the heart of this dispute. Anthropic didn't want its technology used for mass surveillance or fully autonomous weapons. So why was the Pentagon resisting that language?
BRUMFIEL: You know, the Pentagon said this wasn't really about some desire to build autonomous weapons or conduct mass surveillance of Americans. They just didn't like the idea of a contractor telling them how they could use the technology. But Jennifer Huddleston at the libertarian Cato Institute says, these issues are important, and they really haven't been resolved.
JENNIFER HUDDLESTON: We still have not established guardrails with regards to the potential use of artificial intelligence for autonomous lethal weapons or for potential surveillance.
BRUMFIEL: You know, Huddleston points out to the controversy around Flock cameras that can read license plates in communities all over the country. And I note, just this week, The New York Times reported that Russia used a fully autonomous drone powered by an Nvidia chip to kill civilians in Ukraine. So this reflects real sort of very live issues the law hasn't settled yet.
SUMMERS: OK, this is super interesting. Geoff, what comes next?
BRUMFIEL: Well, Anthropic said it welcomed the ruling, and it wanted to continue to work with the government, even on national security stuff. The Pentagon nor the Department of Justice responded to our request for comment, but Mehta thinks they might appeal this, and if they do, it could go all the way to the Supreme Court.
SUMMERS: NPR's Geoff Brumfiel, thanks for keeping us updated.
BRUMFIEL: You bet. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. 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 NPR’s programming is the audio record.
