7 top AI companies sign Pentagon deal for classified work

[0:00] The Pentagon just inked a deal with several big tech companies to use their AI tools. [0:04] Seven in all, including Microsoft, to open AI, Google, and SpaceX. [0:08] But not looped into that deal? Anthropic. [0:11] It's been blacklisted by the Trump administration over the company's demand for the Pentagon [0:15] to include certain safety guardrails when using AI in warfare. [0:20] Let's get more now from CNN AI correspondent Hadass Gold. [0:24] And Hadass, look, this is a very obvious leaving out of Anthropic. [0:29] Tell us more about this deal. [0:32] Yeah, so until recently, Anthropic's clawed was the only AI model that could be used on Pentagon's classified system. [0:38] A lot of the other companies had deals for unclassified systems. [0:41] Anthropic's was the only one that could be used on classified systems. [0:44] Then they had their disagreement over AI safety guardrails. [0:47] Anthropic wanted guardrails around their AI being used in autonomous weapons [0:51] and in the mass surveillance of American citizens. [0:54] They had this disagreement with the Pentagon. [0:55] The Pentagon then labeled them a supply chain risk. [0:57] They are fighting it out in court right now. [0:59] So the Pentagon had to replace Anthropic's clawed with something else on their classified system. [1:04] They started with OpenAI's model, and now they've announced these deals with seven other tech companies. [1:08] Some of them are very well known, like Microsoft, like Amazon. [1:12] Reflection is a smaller AI company. [1:14] It's open source. [1:15] And they're trying to bring in all these companies. [1:16] That's, you know, good. [1:17] You want as many tools as possible for the Pentagon to be able to use. [1:21] But I wanted to point out that there's a bit of cognitive dissonance coming from the Trump administration [1:25] when it comes to what they're doing with Anthropic. [1:27] Because on one end of the administration, you have the Pentagon, [1:30] which has declared Anthropic a supply chain risk, saying that they are essentially a dangerous company, [1:35] and nobody who does work with the military can use their products. [1:38] But on the other end of that, you have the White House, [1:40] which recently had a meeting with Anthropic CEO Dario Amadei, [1:44] meeting with Susie Wiles, the White House chief of staff. [1:46] The White House calling this meeting productive. [1:48] I know that just this past week, or just this week, [1:51] Anthropic was at the White House again to have a meeting about cyber and AI. [1:54] Axios is now reporting that the White House is actually working on guidance [1:57] that would allow agencies to get around the supply chain risk designation [2:00] and onboard their new model. [2:03] So is this company so dangerous that it's a supply chain risk? [2:07] Or was this about politics and a disagreement publicly? [2:10] And I'm sure these issues will come up in court. [2:13] And the reason behind this is just Anthropic makes really good models. [2:16] They're one of the top AI companies. [2:18] And I think a lot of parts of the administration are recognizing that, [2:21] especially when it comes to cybersecurity and the risks that are posed by AI-powered cyber models, [2:27] you want the best possible tools available to you. [2:30] And I think that's why you're seeing a bit of this cognitive dissonance. [2:33] These other companies, as far as we know, [2:34] they have agreed for their AI models to be used for all lawful purposes. [2:39] It seems as though that might have included the issues that Anthropic was worried about. [2:43] Maybe they didn't have those same guardrails. [2:44] And we've seen some pushback from some of their employees, including Google. [2:48] Hundreds of Google employees signed an open letter trying to argue that their executives should not sign this deal. [2:53] One of them even publicly calling it shameful. [2:55] But Google has said that they're proud to sign this deal, [2:58] and they are going to be working with the Pentagon on this. [3:00] Jess. [3:00] Okay. [3:01] More to come on this. [3:03] Hedas Gold, thank you so much for your reporting. [3:05] Just a short time ago, [3:05] the Wall Street Journal reported that the Defense Department has come to an agreement with several top tech companies to use their artificial intelligence in classified settings. [3:17] It comes after Axios reported on a meeting between the Army and some of those companies. [3:22] Sam Saban of Axios broke that first story, and she joins me now. [3:26] Sam, it's great to see you. [3:28] Can you explain maybe to our viewers who are seeing this as breaking news? [3:31] The Pentagon has agreed to use some AI in classified settings. [3:35] Why that's news and why that's important? [3:39] Yeah. [3:39] So I think for those who are blessed to not be thinking about this day in and day out, this may be a shocking. [3:45] It seems like the Pentagon is signing a deal with literally every major AI company that it can find to use their tools and classified networks. [3:54] But this has been part of maybe a month long since February, back and forth with the Pentagon. [4:01] And it started with Anthropic, which is the maker of Claude AI, where there was a huge fight between the Pentagon and Anthropic about how to use Claude in classified networks. [4:13] Anthropic didn't want their tools used for things like mass surveillance. [4:17] The Pentagon was concerned about an AI company setting terms on how it can be used when they have laws and authorities that dictate how they can even use certain technologies to begin with. [4:29] And it's been an ongoing back and forth. [4:31] The White House is involved, like every department, agency, et cetera, has been involved in this brouhaha that's been going on since February. [4:39] And at the same time, for weeks, we've seen other AI companies think OpenAI, the maker of ChatGPT, or Google, which has Gemini, making side deals with the Pentagon as well. [4:50] And so you reported on this meeting with these AI companies and the Pentagon. [4:56] Meeting about what exactly? [4:57] Was it a discussion or was it an ask for the Pentagon? [5:03] Yeah, so the meeting that happened earlier this week was with the U.S. Army, specifically the U.S. Army Cyber Command, and then their other organization, Cyber Command, which is happening, which has within the Pentagon. [5:17] What they did on Monday was maybe unrelated to this, but kind of underscores how quickly the Pentagon is moving into deploying AI technologies. [5:25] They met on Monday with about 12 or so companies, think the companies I just named, OpenAI, Google, Palo Alto Networks, a major cybersecurity company, and others, to kind of run through how they can be using agentic AI to protect their systems. [5:41] Right. The Army is concerned about adversaries, maybe trying to target their critical systems and shut them down and make them unusable if they're in battle or going through an operation. [5:53] Or maybe, you know, think China, Russia, et cetera, other foreign adversaries, just see an opportunity to target the Army and its capabilities. [6:02] And so there was a tabletop exercise Monday. [6:04] They had this meeting. [6:05] It had been in the works for months already. [6:07] And the Army was eager to learn more, had all these CEOs, top security leaders there to chat through it. [6:14] And then, of course, what, four days later, we are now chatting through this major Pentagon deal that is across the entire enterprise of the U.S. military and beyond. [6:26] Things are moving so, so quickly. [6:29] Sam Sabin, thank you for trying, you know, helping try to keep up with it all. [6:33] Appreciate it. [6:34] After blacklisting AI company Anthropic, there is new reporting now that the White House is actively working on essentially ways to get around their own restrictions. [6:41] This is coming from Axios. [6:43] According to sources, an executive order has been drafted to the Trump administration, a way to, quote, dial down the Anthropic fight. [6:50] You'll recall it was back in February that the administration ordered all federal agencies and military-linked contractors to stop working with Anthropic, [6:58] designating the company a supply chain risk, which was an attempt to punish Anthropic after the company refused to let the Pentagon use its AI without restriction. [7:09] But now Anthropic's new mythos model seems to be too important or too good to pass up. [7:17] Joining us right now is Maria Curry. [7:18] She's the Axios reporter behind this new reporter. [7:20] As I read this, I was just, it was fascinating, the, like, turnabout that is being attempted here. [7:26] But tell me what you're learning and tell us more about what you're learning the White House is considering here. [7:31] And what does it mean for this fight that the Pentagon has already picked with Anthropic? [7:36] Right. [7:36] So this is an effort from the White House to get industry feedback and figure out a way to get around its own decision earlier this year to designate Anthropic a supply chain risk, [7:47] meaning that, you know, no one can use this technology because it is so dangerous. [7:52] This is usually treatment that is reserved for, you know, foreign adversary countries. [7:57] Now the White House is rethinking this strategy. [8:01] It was once fully aligned with the Pentagon on this designation. [8:04] But with the release of Mythos, this model is so powerful in potentially cybersecurity settings in detecting threats and preventing these types of threats and attacks from happening [8:15] that the White House acknowledges this. [8:18] White House Chief of Staff Susie Wiles acknowledges this. [8:21] She got the ball rolling earlier this month in meeting with Anthropic CEO Dario Amadei. [8:26] And now they are figuring out a way to bring the technology back into the government. [8:29] In terms of what this means for the Pentagon dispute, that's going to be a much thornier issue for later this later this year. [8:37] Yeah, definitely not over yet. [8:39] Sorry, there was an audio hiccup. [8:40] I thought you'd finish speaking. [8:41] Sorry about that. [8:42] One. [8:42] I love this one line from your reporting. [8:44] One source described the White House effort as a way to save face and bring him back in. [8:48] I mean, does do from your reporting, do they recognize this is somewhat of a problem of their own making? [8:54] Yeah, I mean, you know, this is definitely a problem of their own making. [8:59] And I think, you know, what's happening now is Anthropic is moving to expand its mythos access to even more organizations. [9:07] And my understanding is, is that this is causing a hiccup in the White House's attempts to bring Anthropic into the government. [9:16] They are worried about expanding access to mythos. [9:20] There not being potentially enough compute power for releasing this to more organizations and what broadening this might mean for the security implications. [9:29] But in terms of what we have been reporting on more this morning, our understanding is that this is not a blanket opposition to Anthropic wanting to release its model more widely. [9:40] The two sides are continuing to talk about this and what the timing of such a rollout should look like. [9:45] And at the end of the day, Kate, you know, the White House does not necessarily have the control to stop Anthropic from releasing this model. [9:52] We're in murky legal territory right now. [9:56] But Anthropic, if it chooses to release mythos more broadly, can do so, is our understanding from this morning. [10:02] All right. [10:03] Clearly much more to come. [10:05] Stay tuned on this. [10:06] It's great reporting. [10:06] Thanks for coming in. [10:07] Thank you.