The Battle Over AI in Warfare

0:05The U.S.

0:06Jessica Mendoza:attacks on Iran have unfolded at unprecedented speed and precision. The Department of War launched Operation Epic Fury, the most lethal, most complex, and most precise aerial operation in history. That's thanks in part to a cutting-edge weapon never before deployed on this scale, artificial intelligence. Defense Secretary Pete Hegseth has stressed the creation of, quote, an AI-first warfighting force. And the tech the Pentagon has relied on most heavily is anthropics. But our colleague Keech Hagee says there's irony in the fact that the Pentagon has used anthropics, Claude, in Iran. So literally hours before those strikes were ordered, the president directed the federal government to stop working with anthropic.

0:54Jessica Mendoza:President Trump also called Anthropic, quote, left-wing nutjobs. The Trump administration is trying to cancel its contracts with Anthropic and to designate it as a security threat that no federal agency can do business with. Yesterday, Anthropic struck back, suing the administration. At the heart of the dispute, which has been going on for months, is a fight over values and how AI can be used in warfare.

1:24Jessica Mendoza:And you've been reporting on this saga, right, between the Anthropic, the Pentagon. How would you characterize the drama here?

1:32Keach Hagey:It's completely unprecedented. I've been asking people around in D.C. over the last week, have you ever seen anything like this, where a vendor to the U.S. government became this public punching bag in this way? And no one can think of an example. And I think it really shows us that AI is a different kind of technology, right? It asks new questions of our society that we just have not worked out yet.

1:58Jessica Mendoza:Welcome to The Journal, our show about money, business, and power. I'm Jessica Mendoza. It's Tuesday, March 10th.

2:11Jessica Mendoza:Coming up on the show, the battle over AI in warfare.

2:26Jessica Mendoza:This episode of The Journal is presented by Intuit Enterprise Suite. If your finance team spends more time finding data than using it, if there's one entity here and one here and one here and one here, if scaling your business feels like starting over, you need the Intuit ERP. Intuit Enterprise Suite, the AI native ERP is here. From the makers of QuickBooks. Learn more at intuit.com slash ERP. This episode is brought to you by Alexa Plus. Say hello to Alexa Plus and see how Alexa can do more for you. Need tickets to that last minute show? Craving your favorite restaurant? Sit back, relax, and talk naturally.

3:09Jessica Mendoza:Alexa's on it. Alexa learns your preferences to create a personalized experience. And now Alexa Plus is free with Prime on your Amazon devices like Echo and Fire TV. Learn more at Amazon.com slash Alexa Plus.

3:29Jessica Mendoza:Can you introduce us to the two men at the center of this fight? Secretary of Defense Pete Hegseth and Anthropic CEO Dario Amadei.

3:39Keach Hagey:It would be hard to imagine two more different men. Pete Hegseth famously is a former Fox News host. He has sort of made his brand as being a critic of wokeness and DEI in the military.

3:55Jessica Mendoza:For too long, we've promoted too many uniformed leaders for the wrong reasons. Based on their race. Based on gender quotas.

4:04Keach Hagey:Based on historic so-called firsts. He has this very sort of brusque take-no-prisoners style. And Secretary Hegseth has made using AI and using the sort of innovation of Silicon Valley as a key touchstone of his strategy for the department.

4:22Jessica Mendoza:Dario Amade, on the other hand, is known for being more of a philosopher. He's a lifelong vegetarian who writes in-depth about AI safety.

4:32Keach Hagey:He's a scientist. He has this curly hair that he sort of twirls when he thinks. He loves to communicate in these long, and I mean long, philosophical tracks.

4:42Jessica Mendoza:If you look at the situations I found myself in and the situations humanity has found themselves in. Like, there's so many times where it's, you know, very hard and there's this enormous suffering. And yet there's also this incredible, you know, this incredible inspiration that kind of, you know...

4:59Keach Hagey:He also sort of believes in hashing out these things sort of in public, right? Let's have a, like, a public thinking through.

5:07Jessica Mendoza:Transparency.

5:08Keach Hagey:Exactly.

5:11Jessica Mendoza:Amade co-founded Anthropic with the aim of prioritizing AI safety over business goals. The company has written a moral constitution into its AI models.

5:21Keach Hagey:So Anthropic was founded by a bunch of dissidents from OpenAI that broke off to create their own company in early 2021. So there is not a lot of love lost between these folks going back a while, and that has only become more pointed as they have become real rivals in the business realm.

5:44Jessica Mendoza:One way to get ahead of the competition? Scoring a contract with the government. In 2025, Anthropics Claude became the first large language model cleared to work with classified material. But late last year, the Pentagon began discussing adding new language to its contract with Anthropics, language that would allow the company's tech to be used for, quote, all lawful scenarios.

6:07Keach Hagey:And this really came down to something called a usage agreement, right? They didn't want anything in the Pentagon's usage agreement with the tech providers that tied their hands in any way. And it was really that usage agreement that was the sticking point between Anthropic and the Pentagon. What is at the heart of that dispute, would you say? At the very heart are two issues that Anthropic says are its red lines, that it is not willing to have its technology be used for. That is autonomous weapons and mass domestic surveillance. And these are two things that Dario has written in his essays he thinks would be very problematic uses of the technology that his company has been developing.

6:51Keach Hagey:And that were the sticking points in the negotiations with the Pentagon. And ultimately, it was the hill of mass domestic surveillance that Anthropic chose to die on.

7:01Jessica Mendoza:There, we're worried that, you know, things may become possible with AI that weren't possible before. An example of this is something like taking data collected by private firms, having it bought by the government, and analyzing it en masse via AI.

7:17Keach Hagey:Are those practices legal? The issue is what this whole fight has shown is that the law has not caught up to the state of technology. So we're learning that a lot of what we think the government can't do, it's not because it's illegal. It's just because it can't technologically do it. And like, for example, if we think about the Snowden revelations, right? Edward Snowden, the 29-year-old CIA contractor responsible for what's being called one of the biggest intelligence leaks in U.S. history. Among the things that were so controversial in those revelations were the call logs, right, between people.

7:57Keach Hagey:That was something the NSA had been collecting. And it was a major revelation that, oh, my gosh, every single, you know, call to and from is being collected by the government. But what we also learned is the government has no ability to actually look through that and make any sense of it, right? It's just too much data. Now with AI, all of a sudden, things like that might be legible in a way that they were not before. And I believe the folks at Anthropic believe we do not currently have the laws in place sufficiently to protect us against stuff like that.

8:31Jessica Mendoza:And because of the questions around the legality of it all, Anthropic wanted its usage agreement with the Pentagon to specify in writing that its tech wouldn't be used for fully autonomous weapons or for mass domestic surveillance.

8:44Keach Hagey:And the Pentagon said, who are you to tell us how to use your tech? And I think this was really about setting a precedent for the future, right? They just did not want this concept that the tech companies could tell them what to do to even be in the conversation.

8:58Jessica Mendoza:The Pentagon gave Anthropic a deadline to come to a new agreement, without Anthropic's redline exceptions. That deadline was Friday, February 27th. If Anthropic didn't agree, the Pentagon would cancel their$200 million contract. And the Pentagon didn't stop there. They also threatened to label the company as supply chain risk, which could limit Anthropic's ability to work with any companies that also work with the Defense Department. That could torpedo many of Anthropic's business prospects. As the deadline loomed, another AI company stepped in. OpenAI, the maker of ChatGPT and Anthropics' longtime rival.

9:37Jessica Mendoza:And OpenAI CEO Sam Altman and Anthropics Dario Amadei haven't exactly been the best of friends.

9:43Keach Hagey:There are a couple of moments in the last couple of months that really crystallized their relationship. One of them was the Super Bowl ad that Anthropics ran against OpenAI, which didn't say the word OpenAI, but it was very pointedly against the fact that OpenAI had announced they were going to do ads. And then, of course, there was this epic moment when they were in India together and everyone was joining hands. And those two guys were on stage and they just would not hold hands. But I think that tells you everything about their relationship, right? It's gotten more and more tense. And so what Sam Altman did was begin talking to the Pentagon.

10:21Keach Hagey:And in fact, what Sam Altman said was, OK, let's talk about that. But also, these threats that you are hurling right now at Anthropic to label them a supply chain risk are very bad for the country and bad for the industry. So those are like not appropriate threats. Please don't do that. We can find another way.

10:42Jessica Mendoza:On Friday, February 27th, the same day as Anthropic's deadline, Sam Altman announced that OpenAI now had its own deal to work on classified material with the Pentagon. OpenAI actually says it has the same red lines as Anthropic. They don't want their models being used for mass surveillance or autonomous weapons. But rather than rely on the usage agreement to enforce those red lines, OpenAI said they'd build protections into the tech.

11:08Keach Hagey:OpenAI tried to approach it technically, saying that they would build a safety layer that would make it so that the model just wouldn't be able to do bad things if you asked it to do them. So rather than have a usage agreement that said no domestic surveillance, they would put a safety layer that would somehow make this impossible. What was the reaction to OpenAI's deal with the Pentagon? The company definitely took some heat for it, for the timing seemed opportunistic, and so much so that, you know, Sam Altman came out and kind of apologized for the timing, said it did look opportunistic and sloppy.

11:47Jessica Mendoza:At Anthropic, Amadei wrote an internal memo addressing its rival's deal.

11:53Keach Hagey:Dario Amadei called those things safety theater. In his scathing memo sort of responding to this whole moment, they were— sort of think that that's not enough. They think that part of the problem is that, okay, maybe the laws right now don't allow mass domestic surveillance, but there's nothing to keep Secretary Hegseth from changing the laws or changing the interpretation of the laws in the future, and they wanted something more ironclad.

12:20Jessica Mendoza:Amadei also said in that memo that the Pentagon is targeting Anthropic because it hasn't, quote, given dictator-style praise to President Trump. After the memo leaked last week, Amadei apologized for what he wrote in it. But the Pentagon had already officially designated Anthropic as a supply chain risk. Secretary Hegseth was true to his word

12:42Keach Hagey:and did designate them a supply chain risk. It means that the government has determined that it's not safe to use your technology and that no entity in the Pentagon is allowed to use it. Pentagon's really, really big. So that's a lot of potential customers that you lose.

13:01Jessica Mendoza:The administration said agencies would have six months to transition to other AI models. The move could have far-reaching consequences for Anthropik partners and investors, especially those who are also government contractors, including Lockheed Martin, Google, and Microsoft. What that could mean for the future of Anthropik's business is next.