Taylor Lorenz’s Power User - Inside The AI Scandal Rocking Silicon Valley: How The Govt Wants AI To Kill Without Humans

Episode Date: March 6, 2026

The AI industry just had one of its biggest controversies ever.Support my independent journalism:🙏 Patreon: https://www.patreon.com/cw/taylorlorenz      🗞️ Substack: https://www.usermag.co      A massive conflict between the Pentagon and leading AI companies like Anthropic and OpenAI has triggered a nationwide debate about how artificial intelligence should be used in warfare and surveillance. What started as a $200 million government contract fight quickly spiraled into a Silicon Valley crisis involving autonomous weapons, mass surveillance, and the future of AI regulation.In this episode of Free Speech Friday, I sit down with The Atlantic's Ross Anderson to uncover the dark truth behind the Pentagon’s push for fully autonomous weapons and mass domestic surveillance. We discuss whether AI models like Claude and ChatGPT should be used for military operations, drone targeting, and intelligence analysis. The controversy intensified when Anthropic reportedly pushed back against certain Pentagon requests while OpenAI later reached a separate agreement with the Department of Defense.We break down exactly how negotiations between Anthropic and the Department of War broke down over using AI to analyze the commercial data of American citizens. From Pete Hegseth going "nuclear" and labeling Anthropic a "supply chain risk", to OpenAI and Sam Altman swooping in at the last minute to claim the government deal, this is the AI drama that broke tech Twitter. Plus, we discuss why Claude experienced a massive subscriber spike while ChatGPT took the military contract.Topics covered in this video include:• The Pentagon AI contract controversy • OpenAI vs Anthropic drama explained • Autonomous weapons and AI warfare • AI surveillance concerns • Silicon Valley reactions to the deal • The future of AI in national security

Yeah, it's super interesting that they've had this huge spike in subscriptions. So the point where yesterday, Claude went down like several times during the day because people were using it so much. Over the past week, what started as a contract dispute between AI company Anthropic and the U.S. government has devolved into a chaotic split that is upended Silicon Valley, thrown the government into chaos, and kicked off a nationwide debate about the ethics of AI use in warfare. The debate has centered around two main use cases of AI and whether Anthropic and Open AI, two of the most powerful AI companies in the market, would allow their tech to be used by the government for those use cases. The use cases were the development of fully autonomous weapons and mass surveillance of U.S. citizens. Ross Anderson is a staff writer at the Atlantic and he's been covering this controversy. He recently issued a report that revealed new details about the types of surveillance open AI might have a great. agreed to undertake on behalf of the government.

Today, he's joining me to break down the blowup happening in Silicon Valley over all of this and explain why this could be a watershed moment when it comes to how AI is used by the government. Hi, Ross. Welcome. Tay, thanks for having me on. Longtime listener. I know. For people that don't know, we used to work together. We did. What a lucky and two short season of our lives. So you wrote this great article in the Atlantic titled Inside Anthropics Killer Robot Dispute with the Pentagon. and you have a lot of really great reporting on kind of just the chaos that has unfolded over the past week.

But for people who haven't been following, you know, this tech drama as closely, can we back up and talk about how all of this started? Yeah. And I think probably the place to start is last summer when Ethropic actually signed this deal. They signed this $200 million contract with the Pentagon to basically allow their models work, as people have been jokingly call it, to be used on classified system. by the Pentagon and other intelligence agencies and what have you. So far as my reporting goes and the reporting of others has gone, it seems like everybody was happy for a while, particularly the Pentagon. Like I think at the commander level, from what I've heard, people like many of us, were quite delighted with Claude and preferred it to some of the other models.

Correct me if I'm wrong. But didn't the Pentagon essentially award anthropic Google OpenAI and XAI all these government contracts that were up to $200 million each to develop AI systems that could be used in national security. Yes, that's correct. And then Anthropic, it seems like, was just the first to obtain actual government clearance because their model was seen as, I guess, the most reliable and the safest. I just think it's funny because throughout all of this, Elon Musk has been tweeting about

XAI. It's like, even the government is like, no, we don't want to use it. Please stop. It is not ready. I saw such a great tweet about that where someone was like, he wants the Grock Stormtrooper so bad, but the Pentagon's like, your model sucks. Yeah, literally. But I mean, it seems like in addition to having kind of like the safest and the most reliable model, the Pentagon also initially agreed to some restrictions where essentially Anthropic was saying

that Claude could not be used for fully autonomous weapons or mass domestic surveillance. Yes, yes, that's my understanding as well. And there's kind of a mix of reporting of reporting on how the rupture came after that, right? Because the Pentagon signed that deal and seemed okay with it. And it doesn't sound like Anthropic was all up in their business on a daily basis being like, oh, you can't do that, you can't do that, you can't do that. It seems like the relationship had kind of gone smoothly along for many months. It seems like the turning point came when the United States leveraged Anthropics technology to conduct the raid on Venezuela and capture Maduro.

According to reporting in Axios and the Wall Street Journal, Anthropics technology was used through its partnership with Palantir, and some executive at Palantir told some executive at Anthropic, basically like, shout out to you guys, we used your tech to capture this world leader and maybe kill a bunch of people and, like, bomb Venezuela. And then words sort of made it back to anthropic leadership, and they were like, wait a minute, what? They reached out to the Pentagon and were like, hey, you guys wouldn't be, you know, using our technology in ways that we didn't agree to.

And then Pete Hexeth, it seems like, just went nuclear in response. Yeah, that is what I've heard as well. And one of the things that's yet to be resolved is whether this was kind of a clash of principles or whether it was a clash of personalities. And, you know, from what I've heard, the Hegseth reaction to this was like really, really strong. And that kind of since then, he's had this kind of single-minded pursuit of, like, dominance really, like just not wanting any restrictions and not really caring about the merits of the case, but just not wanting to feel like some tech company can rein him in. Yeah.

It seems like pretty much immediately he went ballistic. And as you mentioned, I mean, this man is a former Fox News host. He's not exactly known for restraint, I guess, in his communication. But he starts this, like, aggressive standoff where, Anthropic where he's like, how dare you question us? And I don't want to like sympathize with Anthropic because they also are, you know, partnered with Palantir and not doing great things. But they're basically like, well, wait a minute, like we just kind of want to make sure

that you're abiding to the terms of our agreement. And we do have these sort of bright lines. And here's the reasons why. And pretty quickly, it seems like things just break down. And before you know it, the Department of War is declaring anthropic supply chain risk, which means that no government vendors across the like Boeing, Lockheed Martin, et cetera, can use Anthropics technology in any way, or they could potentially lose their government contract. This sort of response is something that we normally use for companies like Huawei or companies

that are like a legitimate national security threat, not just like an AI company that was a little too presumptuous in negotiations with the government. Yeah, I mean, there are many Chinese AI companies that are not designated supply chain risks. And it's not just that it's extreme from a security perspective, right? its extreme and its potential effects on Anthropic because, as you know, Google and Amazon and almost anyone who might give them compute that they need to kind of run their business at a really fundamental level, likely do some business with the government or with the Pentagon specifically. And so if there's a supply chain risk, it could be hard for them to find vendors.

You really had some great details in your article about kind of like how this breakdown happened and like what exactly they were disagreeing over. Because I think a A lot of people, they're like drawing chalk, you know, messages outside Anthropics headquarters. And they're like, we love you. We stand you. Resist Anthropic. But it seems like Anthropic was actually quite willing to negotiate here and agreed to some pretty bad stuff. But also the Pentagon wanted some really, really bad stuff.

So can you walk me through kind of what you're reporting found in terms of how the negotiations went? Yeah. The Pentagon were the ones who wanted to unilaterally redo this deal. So like the first sticking point that the first. Pentagon had was they didn't like that these restrictions that Anthropic wanted on the use of Claude in autonomous weapons. How would you define autonomous weapons for people that don't understand what we're talking about?

Yeah. And it could take a lot of different forms, but just technically it's a machine that's empowered to engage targets, which is a fancy way of like shooting at somebody, whether it's an object. Yeah, killing people on its own, making its own decisions in the field. And I think the thing to picture just the way the state of warfare is moving and really is now in Ukraine and Iran elsewhere, I think the thing to picture is a drone, right? If a drone is out in the field and can make its own decisions about who to shoot, that's an autonomous weapon.

So to your point about, you know, whether we think of Anthropic as the good guy in this situation or merely as like the sort of slightly less bad guy, Anthropic was delighted, you know, to have their AI used in these systems potentially in the future, actually had no objection to the idea of an autonomous weapon and in fact thought that maybe like with self-driving cars, which appear now to be, at least in many cases, a lot safer than human driven cars. You know, their theory is kind of like, well,

maybe there's a world in which our killer drones, like kill fewer civilians, you know, and this is something that we should be happy about. But they felt strongly that the model just aren't there yet, that they, you know, risk having the kind of confusion out in the field that could potentially lead to either more civilian deaths or friendly fire. And so they wanted serious restrictions on that. And how did the Pentagon respond?

Well, so one of the things that was floated was the Pentagon was like, okay, great. How about we leave Claude just in the cloud? So meaning it wouldn't actually be in those drones. It would just be in its little data center, maybe synthesizing intelligence, helping with operations, and then any, you know, AI that's running an autonomous weapon is not going to be clawed. Anthropic didn't really like that solution because they reasoned, I think, compellingly, that that distinction really no longer holds the way that these wars are being fought, that all these drone swarms and drones are going to be, if not now, run on mesh networks,

which is basically the way to think about it is trying to push the cloud as close as possible to the drones, if not actually controlling them all the time. Got it. So basically they're like, okay, well, that doesn't really make a difference. So will you please agree not to do this? I think that distinction is so important to make, though, in terms of what you said about like that they're not opposed to the idea of it. They're literally just like it's not 100% ready. And I think that's something that maybe a lot of these people that are standing anthropic right now,

maybe missed. Like Dario even said this in his interview with CBS. He's like, it's just not ready yet. And who knows, you know, also, I mean, there is an error rate, of course, like when it's ready. But I think it's really concerning that they are so eager to actually get there because you can imagine a world where as soon as they believe that they are ready, these type of technology is rolled out without sort of oversight.

Yeah, for sure. And I think if I could speculate on this, I think there's almost like a polarization dynamic happening where people are so. turned off by the kind of like total capitulation of Open AI with Greg Brockman, you know, the president being like the largest funder to the Trump super PAC to, you know, them stripping their non-profits down. It's like it's so clear that Open AI has kind of like is just trying to make all the money in the world and has sort of shed a lot of the principles that at least had expressed early on.

The other thing I really want to underline on the autonomous weapons is that not only are they already kind of in this sort of more of a gradient than a clear line between the cloud and the edge, but also the whole developmental direction that the Pentagon itself and everybody else is pushing is to get the model, like to collapse that distinction entirely. And so I do think it was far-seeing to kind of to not treat that as a solution. In addition to the Pentagon's desire to use Anthropics AI in autonomous weapons, they also sought to use it in mass surveillance. Yeah, the Pentagon One technically is supposed to be about, you know, fighting overseas adversaries or rather fighting other countries, not Americans.

And that's kind of its stated purpose. However, they have this carve out where they're allowed to perform surveillance for other domestic agencies, you know, like for instance, the Department of Homeland Security. And so I think Anthropic rightly had this concern that, as we know, these models are very, very good at synthesizing huge amounts of information and research. it really quickly and generating intelligence from it. And so the concern was that the Pentagon was going to be, through Palantir or whoever else, spying up tons of commercial bulk data and using that to spy an American citizens.

And in fact, per source, you know, very, very close to the talks, they kind of had like come pretty close to a deal on Friday before everything blew up. But like they agreed to kind of remove some of their weasel words and like kind of like legalese language. as it concerned autonomous weapons, but they were a real sticking point for them was this desire to use the model for surveillance, you know, to analyze bulk data. And that's how it fell apart. I was so upset to read this, like, because as somebody that covers, you know, all of this quest to remove anonymity from the internet and to strip us of even more data privacy through things like identity verification, age verification, et cetera. It's so terrifying. And I don't think people totally understand what it means when you say commercial data. You give

some good examples in your piece, but can you explain the type of information that these AI models would be sifting through to potentially identify targets? Yeah. I mean, it could be anything from, you know, GPS data to, you know, Google search results to credit card transactions. Increasingly, I mean, this is still semi-theoretical, especially for kind of the bigger models, but so many people, I mean, I was asking my kid, that basically what you ask your chatbot. And I think, you know, this has been covered elsewhere, But those chatbot conversations feel particularly invasive because just the nature of those interactions that people are having. I mean, just people are just by all accounts spilling so many intimate details into these bots and feeling like they can trust them.

And the prospect of that data being sold and the Pentagon being able to analyze it with the chat bots themselves is just like fully dystopian. It's so dystopian. But also, I mean, I think that we have the government, the Department of Homeland Security issuing hundreds of subpoenas, according to the New York Times to unmask, you know, Instagram accounts and social media pages and discord groups that are criticizing ICE where they are seeking to obtain access to DM history, discord, you know, messaging groups. We saw this effort to identity verify like the internet. And so it just seems like the expressed desire of the government to use technology in these ways at the same time as pretty much everyone in Congress seems aligned on allowing even more of this

potentially sensitive data to be exposed to the government. And, you know, also just commercial actors, it just seems like a recipe for disaster. Yeah. And I think people have a sort of particularly intense fear of this when it comes to Trump because he has these kind of just sort of obviously kind of authoritarian vibes. But this has been a bipartisan pursuit, right, of these technologies and these kind of uses for them. So yeah, no, I find it, I find it really scary. And, you know, you have to wonder, like, because of this flashpoint and because of the sort of attention that's paid to AI and And because of maybe the particular personalities involved, we actually found out about this negotiation.

But there's like every reason to believe that this otherwise would have been, this all would have been quite hush-hush. Yeah. And God knows how much else is going on, right? That's like not exposed because it didn't boil into drama. But to get back to the drama, okay. So basically, Anthropic like puts some very light lines, it sounds like, when did things kind of like fully break? And it sounds like at the same time as Anthropics. is negotiating, presumably in good faith.

Open AI is kind of like backstabbing them and going around and also just being like, actually, hey, don't worry about that. Like, we'll do everything that they don't want to do. What did those talks kind of look like? Like, where did things end up on Friday? Open AI is interesting because they weren't really a major player in this story as we were all experiencing it until Friday. Earlier in the week, you know, I think Altman himself got a lot of good press because I

think he went on CNBC and, you know, sort of expressed that he intended to be. in solidarity, you know, with Anthropic and that he had the same red lines as they did. And I think that was probably partly in response to kind of the news media coverage. And certainly, you know, the discussions internally among the remaining mega idealistic open AI employees. And so I think it came quite as a surprise, you know, when we fast forward to Friday and the Anthropic talks break down. According to my source, it happened when they sent over this kind of final offer where they wanted this ability to search bulk data of American citizens. And Anthropics said no. And then even after Trump, so Trump tweets and it's like, Anthropic, you know, we're done with Anthropic.

You know, we're canceling their contracts. They're out of the government, et cetera. And I think inside Anthropic, nobody thought that was the end of the deal. That's kind of like typical Trumpian bluster. And it actually left a window sort of for further negotiations. But then after the bulk data thing, that's when you saw the Heggseth tweet, where he comes out and labels them a supply chain risk and, you know, really crashes out on X. And you could kind of tell, okay, this this looks like it's over. And then, lo and behold, Sam Altman full of surprises, like what, two hours later or something says, like, congrats, we have a deal.

They had apparently been at least a week. And I actually don't know how long they'd been negotiating, but that didn't happen in two hours. Which is so crazy because like you said, there were all these people, again, this is like why it's so crazy to me, these people that are like standing, you know, whatever AI company is sort of like on the opposite side of the government. Because earlier in the week, people were like, open AI. Like we love that, you know, like it's so great that like they're not agreeing to this. Like, you know, maybe Sam Altman and like, you know, is principled and stuff. And then, yeah, as we saw, I'm sure it was more than two hours. to negotiate that deal.

Like they were clearly working behind the scenes as Sam Altman is like collecting all this good press, like turning around and doing it. My favorite thing, too, is that Sam Altman had to post this statement four times. Three times on Friday alone, he kept getting community noted. And I actually just want to read that community note really quick. So he tweets, tonight we reached an agreement with the Department of War to deploy our models in their classified network. In all of our interactions, the Department of War displayed a deep respect for safety and a desire

to partner to achieve the best possible outcome. Then he sort of posts a bunch of, you know, words that are kind of make it seem like, oh, you know, we're like we basically agreed to the same things as anthropic, but it doesn't seem like they have. And immediately he gets hit with this community note that says government officials have contradicted Sam's claim saying open AI will allow the Department of War to use their models for all lawful purposes. This could mean, you know, basically mass surveillance tools and weapons systems with no human oversight. He posted again, I think, to avoid the community note. He gets community noted again. He posted a third time. And then on Saturday or Sunday, he posted it a fourth time. And then OpenAI issued

their own public statements. It seemed like they were trying to be really vague with the language. Like, you even saw reporters debating this on Twitter. Like, what did they actually agree to? So do you have any kind of idea in terms of like what Open AI agreed to that Anthropic didn't? Yeah. I mean, I have to go on their public statements as well. And I think, you know, one thing they agreed to, glad this is coming back around, is the cloud. only provision on the autonomous weapons. They were apparently quite sold by that. And they had some other finesses about that there

would be like open AI employees that were allowed to like go to the Pentagon and I guess watch. But the thing is you can't watch the AI. It's like there it just that. None of that really made any sense. And I think it was all cloaked around under like like our safety stack. And it was pretty buzzwordy.

But I think what my sources all thought that's just only lawful use, you know, which is why the Pentagon was so delighted to agree to it after they had had this huge blow up, right, with Anthropic. Explain a little bit more of this like all lawful use because I think people think, well, that sounds reasonable. Like it's illegal to spy on Americans, isn't it? Which I think people don't realize. But also we've seen from the Trump administration that they sort of just declare anything that

the government wants to do is inherently lawful. Yeah, that's right. And I think in when it comes to making war and military operations, you know, lawful use is something that's going to get litigated out like, three to five years later, you know, after the dust is settled. So I think Anthropic was right to seek whatever limited restrictions they wanted on it, you know, ahead of time. Let's talk about the reaction. Because if you're on like tech Twitter or whatever, I have never seen so much drama.

It's like these moments where everybody comes out and everyone has to have a statement. And I feel like everyone in Silicon Valley is like posting one side or the other. You had obviously Andresen Horowitz like congratulating open AI and everything. But then you also had people, you know, even at Google and other companies saying, like, this is abhorrent, like Dario at Anthropic is right. Yeah. I don't know if you felt this way, but I don't think we've had a Twitter moment like this for tech since the Open AI like coup and kind of knife fight when when they had that big leadership challenge. The first few data points that suggested to me that it was actually going to be quite a bad news moment for Open AI. Like I was personally revolted by them signing this contract within like a or agreeing to this within a couple hours.

But I didn't know how broadly felt that would be. And still you started, until you started to see like conservative thinkers, whether it's like Dean Ball, who was previously Trump's AI guy, or like Ross doubt that just being like completely outraged. Both by, I think the supply chain designation just because there's like, I think a really in that school of thought of feeling that the government shouldn't be able to just like end to a private company because they, you know, are unwilling to do business with them.

The Libertarians were very much. out about this, I will say. Which was funny. And like in emotional terms too, right? Like it was, it seemed really, Twitter was fun for a few hours. I was really enjoying it myself. And then, yeah, it seemed to really gather steam. And I think, you know, Sam Altman, like, wanted to do like, ask me anything about the terms

of the contract. Like, clearly they were feeling like huge, huge heat. Yeah. The Sam Altman crashed out was just incredible because he couldn't stop posting. Yeah, he did try to do this like, ask me. anything, I'll answer any questions, and then kind of stopped. Like, he answered a few people and then sort of just stopped. And then Open AI comms was like back channeling and trying to, you know, get their message out.

There was also just like so much consumer backlash. Obviously, like the majority of the public is not really on Twitter engaged in Silicon Valley drama. But it seemed to have really broken containment where Claude shot to number one in the app store. You had people like Katie Perry tweeting out like herself purchasing like a Claude, with like a heart around it. And I don't know. I mean, even like I talked to my mom, you know, and like she was talking about it. So it seems like a lot of people sort of got this message of like, oh, open AI caved and anthropics stood strong. Again, this is where you started to see the chalk

drawing outside Anthropics office. Like, we believe in you. We love you. Like resist Trump. And it's like this is a company that is actually excited to build autonomous weapons and currently partners with Palantir. Like let's not, you know, go over. board here. But I, but I wonder if this is, you know, a moment for them because we know, especially between Open AI and Anthropics, specifically, like, they have gone back and forth with each other. You saw Anthropic take shots at ChatGPT and Open AI with their Super Bowl ads this year. And they have been sort of neck and neck in terms of their consumer products. I mean, look, chat GPT was the fastest growing consumer product of all time. But Claude does seem to be catching up.

And I don't know, like, is this potentially an inflection point that? where, you know, it could give them an edge. I think so. And one that, you know, provided the Pentagon doesn't burn their company down, like, it's actually purchased at a reasonably decent price because that Pentagon contract, $200 million in the scope of Anthropics revenue is not that much at all compared to this free publicity they just got. And I think a lot of it, you know, the average person reading even like a five-paragraph

story about this might wonder like, wow, why did the Pentagon like Claude so much, you know, and I do think maybe to, you know, to people who are not like too online like you or I who are kind of tracking sort of how good the models are in real time like in my experience Claude has definitely surpassed chat GBT a lot to the point where like I kind of find chat GBT like annoying to use now by comparison just kind of the writing style and it just doesn't seem nearly as sophisticated and so it is yeah it's super interesting that they've had this huge spike in subscriptions. where yesterday their Claude went down,

was down like several times during the day because people were using it so much. It was down for hours, which is the first time that I've seen any of these models really go down for so long, and people were like furiously posting about it, running to subredits, being like, wait a minute, I just moved my whole life onto this platform. I think it's so interesting too, because like I covered back in the day

the war between Bumble and Tinder where there were these two sort of like seminal dating apps that really define the market. Tinder was the first mover and, find kind of like modern dating. It introduced app-based dating to the public. But then you saw Bumble, I think, like, really overtake Tinder in a lot of ways by positioning itself against it's like, you know, this first mover in the market. And I feel like you're seeing something similar with Anthropic where it's like chaty PT was the first mover. It was so widely adopted, but people

also became very aware of its flaws. Like it's the way that it's synchophantic, like the weird phrasing, the M-Dash is like all of these like corny sort of tropes that I think, were kind of consistent in a lot of early models, but are now 100% associated with chat GPT. And you hear people be like, oh, you know, what is that a chat GPT response? ChatGPT, like it is in some ways, like it's good that it has that sort of brand recognition.

But it also, I think, is bad because it sounds sort of associated with everything bad and sloppy about the AI ecosystem. And Anthropic is seen as sophisticated and now seen as more responsible too. I think Open AI has just made it like fallen on their face a few times, like the ads too, right? I think the. Super Bowl commercials that Anthropic ran were really brilliant because, as we discussed earlier, I mean, I was asking my son yesterday. I was like, how much she used chat TPT? Because he doesn't have

clod yet. And he's like, oh, not really that much. I was like, oh, so do you use, like, when you're want to find something out, are you using the Google search bar, you know? And he's like, oh, no, I guess I mostly use chat chit. I'm like, okay. So, so like, people are putting so much into these chat bots. And as you say, usually you would think that that kind of Kleenex or Uber effect, right, where the first mover brands the whole space, you know, down to it becomes a verb. Usually that would be a plus, but in this case, as you say, it's sort of associated with all the kind of most irritating things about like early stage chatbots.

And that does give Anthropic these huge opportunities where if everyone's looking at their product in recent days, I mean, I think most people, I'm not trying to run ad copy for Anthropic here on your podcast, but like most people I do have that reaction, right, where they use it like, Oh, this, it looks a little more sophisticated. It like kind of like it types and paragraphs. It doesn't do that kind of as much of the kind of binary statements. It's not this. It's that, et cetera. I think of the like Winnie the Pooh meme of like the Winnie the Pooh and then the Winnie

the Pooh with like the little like monocle or you know in the tucks or it's like once you see like AI writing, it's impossible to unsee it for me. Like even when it's written with Claude, it's not, you know, significantly better. But it does sort of avoid some of those problems. And as you said, you know, I think people forget how little 200, you million dollars is to these companies. These companies have like multi-billion dollar valuations. They're raising billions of dollars. Hundreds of billions. Hundreds of billions. So like this PR boost for Anthropic, I think could be totally worth it. I think so too, especially if they see it as a

kind of long game thing, right, where okay, there'll be a new administration or even like hegg sets out in a year and a half. You know, Trump, Trump always caves. Like you can see them getting back in the Pentagon, maybe even like within a matter of weeks. And yet, they, they're going to. got this huge kind of commercial for being so principled for yeah not even 200 million dollars what's going to happen now like is chat chappi phogee just going to be used to mass surveil americans and you know swarm bomb people abroad like is this technology being deployed immediately i don't think so i think it'll take time for them to get a model right that can be used on classified systems for a variety of reasons and i think it's also possible that the anthropic thing will get ironed out

maybe not in the immediate term, but there's no way that the supply chain designation is going to stand up in court, or actually, I should say, very little chance. But in the meantime, the Pentagon hasn't actually caved, right? Like, they are still, they still have these desires to have autonomous weapons and mass surveil citizens. And to me, like, nowhere throughout this entire process have they backed down on any of that. And in fact, it seems like they're just seeking to sort of charge full steam ahead. Yeah. And, you know, as you know, Open AI put out a new announcement last night that they've, you know, now amended the deal, which I thought was so funny because, you know, you saw people like Roon, who's an

open AI employee. That's his sort of pseudonym on Twitter and other kind of open AI partisans complaining all weekend about them being, like that their deal was actually great. And, you know, people were complaining about it. And yet they renegotiated it in response to that pressure. I thought that was so funny. Yeah. Well, the open AI employees were fighting for their lives on Twitter, honestly.

And just the open AI defenders, I felt like it's funny because it's like, I'm You see this with Trump a lot, right? Where these like these true believers will defend anything he does and then he will sort of like pivot and then they just pivot along with him. It seemed like that with Open AI where they were like, no, no, no, it's actually the best deal. It's a totally good deal. Why would we ever change it? Open AI, you know, renegotiates and they're like, oh, we were so smart to renegotiate. Like, you know, see how they listen to people.

I don't know how fast they can get in there, but there's been some back and forth on Twitter already about the terms of their new deal and whether it's any lawful use cloaked in yet another like three layers of legalese and, and, and, and, And I don't know. Maybe it's not like the fact that they had to renegotiate it suggests that maybe some of those restrictions are firmer. But even so, I don't believe it because I think they would be touting the firmer restrictions if they could. And I think the fact that they're not being, you know, sort of overtly shouting out like the more specific language means that they're just, it's all just kind of dancing around the facts. But regardless, none of these companies are slowing down. Like even if they are like litigating the terms for now, as we said in the beginning, even Anthropic is like, we, we have. no problem building these systems. Our models are just not ready for it yet. And I think as soon as these tech CEOs decide that their models are ready, whether that is the case or not, these systems will be deployed. And no one in the government has shown any desire to have any sort of restraint in those

areas. Yeah, I think that's right. And I think it's funny, I was talking to to my source who was close to these talks. And he was saying that these models are getting so powerful, or at least if that sounds too complimentary, so potentially useful in kind of military and surveillance contexts that they expected a kind of DPA issue to come, like that the Pentagon might want to compel production of them and sort of maybe even take over, like something you saw at the Manhattan project. And they were just laughing because they were like, this happened way earlier than we thought, but also it's like the stupidest possible version, you know, probably because Pete X and Emil Michael in these, you know, kind of class.

I'm glad this happened, actually. I know, I mean, it's so much of it is sort of depressing, but I don't think the general public is aware. Like, I wasn't aware of the degree to which Claude is integrated into these classified systems and, per my sourcing, like, deeply integrated. Like, apparently Heggseth, who reportedly has no love for Dario, I would the CEO of Anthropic, nonetheless was like, we need Claude, you know. They just, I mean, despite the orders to sort of cut all use of anthropic technology,

there was reporting saying that they just used Claude in their targeting of Iran. Yeah. I think they're relying on it now operationally, which is crazy, given that it's only been in their systems six months. But as I was just saying, when it came to my own kids, like many people are like that now, right? Well, Ross, thank you so much for joining and sharing all your reporting with us. I appreciate having you on. This is so fun. Thanks for having me on. All right. That's it for this week's episode of Free Speech Friday. If you like my work,

please, please support me on Patreon via the link below or buy a paid subscription to my tech an online culture newsletter at usermag.co. That's usermag.com. I publish a biweekly roundup of everything that I'm reading and seeing online. You can also get my newsletter on Patreon where I produce bonus content and do a monthly Q&A live stream. I basically have almost no advertisers on this channel, so truly every single dollar of your support makes such a difference and ensures that I can keep this series going. I'll be back next week with a brand new episode of Free Speech Friday. See you then.