Trump v Anthropic: The battle over who controls AI

0:00This BBC podcast is supported by ads outside the UK.

0:30business needs in one place, saving you time, headaches, and serious money. Stop paying for missing pieces. Go to odoo.com. That's O-D-O-O dot com to learn more. With the new Schwab Teen Investor Account, teens can gain hands-on investing experience and build positive money habits. It's an account co-owned by you and your teen, so you can monitor and engage with the account while your teen learns how to invest and manage money. Learn more at schwab.com. Welcome to a new war for the Trump administration. This is a domestic war. It's a war with an American AI company. The Trump administration say Anthropic is being unreasonable in the demands it puts on the software it gives to the Department of Defense.

1:15The company says, no, this software is potentially very dangerous. We need to tell you what you can and can't do with it. The Trump administration is now saying effectively we're going to put you out of business. Welcome to AmeriCast.

1:30America's America's from BBC News. You hear that? Oh, I think when I hear that sound, it reminds me of money. We didn't start this war, but under President Trump, we are finishing it. This is a big cover up and this administration is engaged in it. This guy has Trump derangement syndrome. I have four words for you. Turn the volume up.

1:57Hello, it's Justin in the worldwide headquarters of AmeriCast in London, England. And it is Marianna sitting next to Justin in the worldwide headquarters. It hasn't been just the Brits for a while. No, it hasn't been just the Brits, but we're not going to be American free. We are going to be joined by a really excellent guest. We've got a guy joining us in a second or two who was the person who advised the Trump administration on their entire AI policy. And what we're talking about today is him effectively saying the Trump administration are now making serious mistakes and doing things that he regards as deeply wrong.

2:35His name's Dean Ball. He was an AI advisor. He went into the Trump administration. He is now very much outside the Trump administration. And I doubt Donald Trump thinks much of him at the moment. So anyway, we're going to be talking to Dean in a second. Let's start off, though, with Anthropic, because people who haven't heard of Anthropic will probably have heard of an Anthropic product, the main product actually, which is this chatbot, essentially, Claude. Yeah, Claude, exactly. So if you like using any kind of AI assistant where you type in prompts and you get answers back, like Claude is of that genre, will generate images for you, will answer questions, will collate information.

3:13It's Claude that was the first to be used by the US Department of Defence to do classified work. And so that was under Joe Biden. And then obviously it continued under Trump. And when we say classified work, we are not talking about just typing in, how do you capture Nicolas Maduro and it coming up with a session? Can you imagine that with some instructions? Arrive in the middle of the night. Exactly. Give him a water bottle. Make sure he's wearing a fleece. It's a little bit more than that, isn't it? And it's a lot more than that, frankly. And it allows organizations to do things at scale and to interrogate data at scale and to think ahead, as it were, at scale in a way that mere humans can't.

3:53That's the value of AI to any large organization, potentially, but particularly in the defense field, of course, it's controversial. And it's controversial because you are talking ultimately about not just capturing someone, as in the case of Nicolas Maduro, but potentially killing them as well. Yeah, that's the best way of thinking about this technology is the way that it can en masse interpret and analyse and collate things that as human beings, it would take us so long to get through and look at, you know, like data being able to really accurately and quite effectively summarise and give you details, but also things like analysing satellite imagery or analysing, you know, which when it comes to war or capturing material or whatever that might be, is generally very useful from an intelligence gathering point of view.

4:37And so it makes sense that AI companies have expanded into this area and these kinds of collabs are happening. And just to be clear about what the row is with Anthropic, it's really two things, isn't it? Number one, should the AI be used, can it be used to surveil in a mass way American citizens going about their business? And number two, should it be allowed, can it be allowed to kill people without a human being being in charge? Yeah, and that was in the original contract that Anthropic had with the department. Those were like red lines that were set out. They were clear. Those were two things that right now they're kind of not comfortable with being a part of or involved in.

5:17And the reason that this row has happened is because last month there was this change in position. So the administration, more specifically Pete Hegseth, the Defence Secretary, basically came out to say, can Anthropic drop these restrictions? They stopped saying that these are red lines and they accept, quote, any lawful use of its tools. And because Anthropic then turned around and said, no, we're not doing this. We are a responsible AI company and our red lines remain. It's now got all a bit messy. Well, to put it mildly, yeah, I mean, all a bit messy, but also in a way, and we're going to hear more on this with our guest in a second, but in a way that is, I'm not saying Anthropik is about to go out of business, but in a way that is really deeply damaging to Anthropik.

5:59So there's now another argument, which is, hang on a second, whatever the nature of this row, how is it that the Trump administration, just because they disagree with a private company, is able to put the frighteners on them, put the squeeze on them to this extent? Well, for example, so Hegseth last month posts on X. Effective immediately, no contractor, supplier or partner that does business with the United States military may conduct any commercial activity with Anthropic. This is off the back of them saying we don't want to drop these red lines. Anthropic will continue to provide the Department of War its services for a period of no more than six months to allow for a seamless transition to a better and more patriotic service.

6:34Pay attention to that adjective. And then you've got Trump on True Social also last month saying the left wing nutjobs at Anthropic have made a capital's disastrous mistake trying to capital's strong arm the Department of War and force them to obey their terms of service instead of our constitution. So they've basically said they're a supply chain risk quote and we're not going to do business with them and you kind of shouldn't either, which is like, whoa, whoa, whoa. Like it feels like it went from zero to 100 very, very quickly. And normally that phrase, supply chain risk, is used about basically companies, foreign companies that they think are an enemy to the United States and could be doing damage to the United States.

7:14It's that level of seriousness normally. I mean, the other context here, and I think it's important, is the likely reason for the change from Pete Hegseth, for example, in the department is because of what's going on in the world right now. Like the strikes on Iran, you know, and the feeling that actually this will increase our capability when it comes to some of these conflicts or some of, you know, some of the military action we're engaged with. And so you get that real sense of, I don't know, my read certainly of their reaction has been, A, that it's all been very politicized. The suggestion that Anthropic are kind of quite deliberately disagreeing with the government because they have a different political approach to Donald Trump.

7:53Their politics are different, which Anthropic would say that's not the case. This is about morality and being responsible. but also Anthropic have kind of positioned themselves as the sort of not Trump AI company to some extent because of this, which is another tactical question that's kind of interesting. But then also you've got like, you almost have that sense of urgency that they're kind of like, hang on a second, we need you to do this now. It sort of feels like when all the tariff stuff was happening, you know, it's kind of like, okay, now you must do this now. And if you don't do it now, then there are going to be consequences.

8:20Right. To an expert and a former insider, Donald Trump's former AI advisory helped to devise a lot of the Trump administration's AI policy. Dean Ball, thanks so much for joining us. Thanks for having me. Now, you have had an argument with the White House about Anthropic. Explain to us from the start, as it were, what it is that you think that they've done that they shouldn't have done. My dispute is actually not with anything the White House itself has done. This is sort of a bureaucratic point. But my dispute is actually with actions taken by the Department of War, Because the Department of War, they had a dispute over contractual terms with the AI company Anthropic.

8:56Anthropic would not agree to their terms. And rather than simply canceling the contract that they had, the Department of War has decided to go on a completely counterproductive and quite possibly illegal regulatory rampage against Anthropic. designating them a supply chain risk, which is a regulatory mechanism that heretofore has only been used on foreign adversary companies by the Department of War, saying this technology is not just something we don't want to work with. It's actually a national security threat. And this is a direct attempt to destroy the business of anthropic. OK, and we're going to get in a minute to what that says, the wider picture of what that says about the way the Department of War works, the way Pete Hegseth, the war secretary, works, etc.

9:41First, though, am I right in thinking it's two fundamental things that Anthropic has not agreed to? One is autonomous lethal weapons. That's simple enough. That is a weapon that is capable of identifying, tracking, and killing human targets with no human oversight. So we're talking about machines making the autonomous decision to kill people, basically. And Anthropik's position on that seems to be something to the effect of one day this is going to be necessary, but right now the AI systems they produce in their assessment are not good enough. This is similar to like if you made a fighter jet.

10:19And you said, yeah, like, we think this fighter jet can fly at this altitude, but if it goes above this altitude, we think it might break apart. And so don't fly the jet above this altitude. It's sort of more like that. It's more of a technical limitation. And then you have domestic mass surveillance, which is somewhat more complicated. Basically, this relates to the fact that in American national security law, surveillance has a particular definition, which in many cases could exclude use cases like this. Say the government buys commercially available data that contains potentially sensitive details about Americans.

10:56So say smartphone location data. Lots of companies sell smartphone location data of millions of people. That's a product that you can buy. Whether or not you should be able to buy that product at all is an interesting question. Surveillance camera data, all sorts of stuff. If the government directly put a chip in your phone that tracked where you went, that would be surveillance and it would be totally illegal under U.S. law. If the government buys data from someone else and then analyzes it, that's actually not surveillance. That's the analysis of commercially available data. And so that's what Anthropic was trying to prevent was basically this thing where there are legal technicalities where something that I think we would probably both consider to be surveillance, like in vernacular terms.

11:37is not actually considered surveillance. And so Anthropic wanted a somewhat more, a somewhat broader protection against domestic surveillance in the contract. And what do you make of those red lines? Do you think that they're reasonable, not reasonable? Do you understand why they've decided to sort of put their foot down on these issues? I totally understand why they've decided to put their foot down on these issues. And it seems entirely reasonable to me. To be fair to the Department of War, it's also reasonable that they say, hey, look, we decide when autonomous lethal weapons are ready for primetime, not some guy in San Francisco.

12:12Now, I think that's actually a little simplistic because in reality, the Department of War and its commercial vendors decide when technologies are ready for primetime collaboratively all the time. These restrictions kind of feel like public policy. They kind of feel like it should be a law that Congress passes. And so like maybe we shouldn't be doing them through contractual terms is a point that I get. I think there are all sorts of interesting things you can say about the principle that the Department of War is basically laying down here, that they should be able to use technology for all lawful use.

12:42There's all kinds of interesting things you can say about that. I've chosen mostly not to interrogate that side of the argument. And I've mostly chosen to say, you know what, I'm just going to assume their principle's great. And I'm going to say I agree with the principle's fine for the purposes of what I need to talk about. Their principles are fine. The problem is the punishment. Do you have a sense then of what it is that Anthropic as a company could have done differently? Was there a way out of this, a kind of compromise? Yeah, they had a variety of options in front of them. And this is why it's hard for me to defend them entirely.

13:17Because look, number one, if these red lines were so important to them that their violation is something that they're like willing to stake the entire company on, then I think one thing you could have said is we're not going to sell this technology to you until there's laws passed that prohibit XYZ stuff. That seems like one thing you could have done. No one else was chasing after this business, right? Like the models weren't even really good enough for a lot of this stuff until very recently for a lot of the military uses we're talking about here. The other thing, you know, and this is more of a realpolitik aspect of dealing with this administration, is that, you know, this administration is a very they make pretty stark distinctions of friend and enemy.

13:55There are all sorts of things that anthropic leadership could have done to go to the Oval Office and make friends with the president before any of this happened and not antagonize the administration. They've done things that have antagonized the administration. And like I was in the administration and when I was in the administration and they lobbied against things that the administration supported or they hired all of the like the key architects of the AI policy from the Biden team. Right. The president's rivals. Right. Like right when the Trump took office, anthropic leadership. was hiring like all of the architects of the Biden AI policy.

14:28That was antagonistic. A couple months ago, they gave a very large sum of money to a political action committee that essentially works to elect Democrats. Right. That's interesting. So the bosses of Anthropic have not done anything to cozy up to the Trump administration. And meanwhile, their rivals at OpenAI are walking straight in, aren't they? Well, let me just say for a second that like, yes, that's true. I'm very reluctant to say that should be a fatal flaw because it shouldn't be, right? Like you should be able to do business with the government and also not support the government politically.

15:00You should be able to support the opposing political party while also doing business with the government and expect not to be harassed by the government. That is actually an expectation that we should have. Now, a lot of people would tell me I'm naive for that and I would tell them that they are nihilistic. Well, yeah. I mean, you're not naive. You're a bit old fashioned, maybe. I don't know. What do you think? Yeah, it's a little old fashioned, for sure. Isn't this slightly the point, though? I mean, when you talk about deciding to basically make an enemy of the administration when it comes to Anthropic, is that not, and I'd be really interested to know what you think of this, like in part, a little bit tactical, a little bit deliberate when you've got other companies like OpenAI that are seen as being much more aligned with the administration.

15:40There are also lots of people who don't like the administration very much, don't like Donald Trump. And could that be commercially wise or unwise? Depends what you think of it, but do you think that it could be deliberate in that sense? Well, I mean, OpenAI is such an interesting example because, first of all, I don't think OpenAI was really trying to swoop in and like hurt Anthropic when they made the deal with the Department of War. I actually think they were trying to be de-escalatory and I kind of defend them on that front. I don't think what they did worked particularly well, but I do think that they had good intentions.

16:10I don't think they were like cynical or evil. At the same time, like it's really interesting because OpenAI has publicly said that they have the same exact concerns as Anthropoc, domestic mass surveillance and autonomous lethal weapons. There are other regulations that have passed that have annoyed the administration, which Anthropoc supported, which OpenAI also supported. These are not military regulations. These are domestic policy regulations that were passed in states like California, where, you know, AI regulations that the administration opposes that both Anthropoc and OpenAI have supported.

16:41OpenAI staff have been vocal on the internet about not supporting this administration. And OpenAI has also said, you know, that one of the big red lines the administration has claimed that they have is we don't want operational restrictions. In other words, what we don't want is for one of these California companies to be able to take away our critical AI services while we're in the middle of a combat operation. And they claim Anthropic has threatened to do that. Anthropic says that isn't true. OpenAI has said that their whole contract hinges on operational restrictions, right? In other words, the model refusing to do things in potentially real time that the DOW wants them to do, wants the model to do.

17:21So all this is to say that if you actually look at the substance of where OpenAI and Anthropic are, they're really not that far apart. But the reason it feels so different is that OpenAI has done a much better job of genuflecting to the administration in ways that, frankly, I think Anthropic leadership, and specifically in the person of Dario Amadei, they have too much dignity to do. They have too much pride. This then brings us to your wider point, which is such an important one, and we ought to get to before we let you go. You're basically saying what the Department of War is doing is, in effect, trying to put Anthropic out of business.

17:57And in a nation where there are private property, private companies mixing occasionally with the government, occasionally deciding not to, that that is something that, what are you saying? We just haven't seen it before. It is of such magnitude that it's a Trump administration thing that we all ought to be noticing more. I would say it's not unprecedented in the history of America, right? You can go back and you can look at Franklin Delano Roosevelt, who did stuff like this. And by the way, when it happened, people like me, aka Republicans, we called it tyranny back then, 100 years ago. It happened during the Nixon administration.

18:32Nixon was a Republican. And Democrats called it tyranny at the time. I mean, if I had been alive, I would have called it a tyranny then. And, you know, a lot of people in the Trump administration, by the way, like specifically admire Richard Nixon and want to emulate the presidency of Richard Nixon in many important ways. So it's not surprising that we're seeing this kind of a parallel. I think there are a few things that are different. First of all, it's the duty of people who observe the tyranny to call it out. And so that's why I'm doing what I'm doing. I don't think it's unprecedented necessarily, but it is necessary to call out as being like, this is a real abrogation of private property and maybe even speech.

19:05If it is indeed politically motivated, this is also an abrogation of anthropic speech rights. One last question for me, but just about safeguards in particular. I mean, something that struck me that you said was about, you know, essentially the Department of War saying, well, look, we're the people that get to make the call about some of these things. But actually, as you acknowledged, there's often a gulf in understanding or expertise between, say, the politicians or the people working in the ministry and then the people who work for the companies who understand the technology. Doing my job, what I kind of have found is that it feels like lots of people just really do not understand the limits of AI or what it's capable of right now.

19:42Like, are we in a good place in terms of keeping people safe and using this technology ethically and in the right ways? Or are we in a place where the technology has moved much faster than any of the politicians understand it? We are in the early days. And so the technology is combusting a lot of our old assumptions that are baked into our laws and institutions. And so I think there's going to be a period of experimentation and we will probably see people do things that we don't like and that we want to create rules against. And I think that's fine. I think that's all well and good. You know, in terms of expertise, I have to say I'm a little disappointed because one of the things that the tech affiliated people with this administration of whom, you know, with whom I worked and in some sense, many people would consider me to be a part of that group of the sort of new tech writer.

20:26I'm really not. I've been in the right for a lot longer than new. The idea was like, oh, we're going to bring technical expertise. These these eggheads in the Biden administration don't know. So we're going to bring people with business experience from Silicon Valley who are whatever, whatever. I think you can go and see things that certain administration officials have said in public that betray a pretty serious lack of technical understanding, not ethical, not moral, but technical understanding of what's going on here. And that's because having worked in Silicon Valley doesn't mean you understand AI, right?

21:00Like, look at all the major—there are tons of big companies in Silicon Valley that are kind of adrift when it comes to AI strategy, right? So just because you have an experience as a technologist doesn't mean you understand AI. And I think basically two years ago, I spent a lot of time trying to explain to policymakers the basics of how this technology works. And I feel I still do that today. Dean, and you've been doing it to us as well. And you've explained that particular situation between Anthropic and the Trump administration brilliantly. Thank you so much for talking to us. Thank you very much for having me.

21:30Great to chat. I thought it was particularly interesting there where Dean was explaining how, you know, ultimately he still feels like he's having to explain how AI works to policymakers and other people involved in this. And actually, you get to this point where it feels like there's a real tension between politicians and the experts in AI, because I guess the technology itself is not inherently political. But then you've got people like Pete Hegseth, for example, at the Department of War kind of saying, look, having guardrails or having restrictions is in and of itself like somehow woke or left leaning, not right leaning.

22:05Whereas actually what Dean was saying to us was kind of the opposite. Like you do just have to test stuff out and have some guardrails. That's not that's not inherently political either way. That's just how the evolution of technology works. And this was what Pete Hegseth said when he was talking about this new strategy to secure American military AI dominance. Today I want to clarify what responsible AI means at the Department of War. Gone are the days of equitable AI and other DEI and social justice infusions that constrain and confuse our employment of this technology. Effective immediately, responsible AI at the War Department means objectively truthful AI capabilities employed securely and within the laws governing the activities of the Department.

22:52we will not employ AI models that won't allow you to fight wars. We will judge AI models on this standard alone. Factually accurate, mission relevant, without ideological constraints that limit lawful military applications. Department of War AI will not be woke. It will work for us. And the point that we were hearing from Dean is that it's not really about woke is it it's about those two things number one can it take part in mass surveillance which as dean said is probably illegal in america although there are questions about not very publicly popular and not very publicly popular regardless of the kind of semantics of what that means yeah and the other thing whether it can kill people without having a human in charge i'm not saying it's necessarily particularly woke to think through the morality of war indeed there are scholars who have spent entire lifetimes and serious scholars of war and serious soldiers, actually, more to the point and perhaps more to Pete Heggs' liking, serious soldiers who think a lot about the morality of war.

23:59And just to sort of dismiss it like that and dismiss the possibilities of AI, it just seems to me either he doesn't fully understand how they might sort of intersect with the law or perhaps he just doesn't care. Do you know what I think? I think that there is a tendency from people like pete hegseth and the administration to presume that ai technology and the companies are similar to the social media companies and if you think they've had this whole battle with the social media companies around what the administration would call kind of censorship or feeling like certain posts are removed and that there's a kind of woke approach to online safety and fundamentally like social media platforms like metas ones instagram facebook TikTok are very different beasts really to AI technology and its kind of raw state.