Extended interview: Trump AI adviser David Sacks on tech leaders' AI fears

Thanks for having me.

Yes.

If the prediction is that AI is going to end humanity, I think that's overly dramatic and I don't personally subscribe to that. I think this is becoming a panic. We're in an election season and there is an incentive for political actors to try and amplify the fear. And I think this is a moment to really calm down a little bit and look at the issue rationally. I think that AI safety is an issue as the leaders of these labs have said, but it's something that I think we can address and it's something that these labs themselves have to get right.

Well, to me that's a little bit of a contradiction when I hear something like that. I mean, you have the leader of a frontier AI lab say that frontier AI is dangerous, so dangerous that it can end humanity. I guess my question when I hear something like that is why are you doing it? I mean, you're the leader of this company. Why are you acting like this is something that you can't control? If you can't control it, then don't do it would be my view. Or step aside for people who can. You have a responsibility as a leader of this company to make sure that your products are safe. You have massive product liability.

Well, it depends what they do next. I mean if what they're saying is they're going to prioritize safety of their products and I think that's a good thing. But I think scaring the public in this way is not good. I think that what a leader should do is say here's what we're doing and why and here's how we're going to make our products safe and this is why you don't have to fear that we're going to do something civilization ending. So this is really on Dario and these other frontier lab leaders to tell us how they're going to do this safely.

Well, I think that what they're saying is that we're at the stage of the development of this product that we should be prioritizing safety or you could just call it the predictability, reliability, robustness of these products. They're already very powerful. And so I think we're at a phase now in the development. We're not stopping development. I don't think anyone's saying that they want to completely end development, but what they're going to do is prioritize the reliability. This is what customers want, by the way. I mean, I don't think we have to interpret this as some sort of altruistic sacrifice on their part. These are business people. I think it makes all the sense in the world for them to prioritize giving customers what they want. No enterprise, you know, who's a buyer of AI wants an AI model that's going to leak their data onto the internet or, you know, inadvertently hack a competitor. Those are not things that we want from our AI programs. So, people want these AI models to be predictable. They want them to be reliable. It makes sense for the labs to focus on that. And I think that's what they're indicating they're going to do.

I think there is a well-known phenomenon known as regulatory capture where this has happened throughout business history and is very well documented where companies will go to the government to seek regulation and try to bend the regulations to their advantage. I do think that is a concern here where you have these companies that are very well-funded, very well established that are on the brink of having a multi-trillion dollar IPO go to the government in order to get regulations that benefit them. We need to make sure that doesn't happen. I'm not saying that I'm against all regulations, but I am against regulatory capture. Let me give you an example. In his essay, Dario says that he needs antitrust waivers in order to pursue this AI safety agenda. I don't think that's true. I don't think he needs what he's saying when he says I need an antitrust waiver is I should be able to form a cartel or I should be protected from competition. In my view, that's not acceptable. You don't get to be protected from competition. There are thousands of companies, millions of companies across America who know they have to make their products safe and they don't get protection from the government for doing that. So look, if what they're seeking is a liability waiver or again an antitrust waiver, I would say no way to that. You have an obligation to make your product safe and you should not need anything extra from the government in order to do that. You already are liable. Lina Khan, who is the former Democrat FTC chair, just came out, I think, with a powerful statement saying that AI companies are already liable for product defects. If you put out a product that's unsafe, you can be subject to civil lawsuits. You can even be subject potentially to criminal liability. And she's right. She's saying something that I've said for a while, which is these AI companies are already subject to lots of laws, lots of regulation, lots of statutes. Again, they have an obligation to make their products safe. And I don't think they should be getting again some sort of special cartel treatment from the government in order to do what they already should be doing.

I think that's part of their motivation and again I think it's a big part. I think that again asking for an antitrust protection so you can form some sort of cartel with your competitors to me that's completely out of bounds. By the way, we already had right after Dario's essay came out we had basically former DOJ officials say that no we actually have policies allowing tech companies to share information and data on cyber security. You don't need an antitrust exemption to do what you're saying. So again, you know, if they want to focus on safety, if they want to prioritize that as part of their product roadmap because they see that as the way to make their products more reliable, better, safer, then that is their decision and no one is standing in their way. You know, one of the things that Dario is saying is that we need to pace the frontier.

Slow it down. But here's the thing that I find curious about this. He is the frontier. He and Sam are the frontier. You know, when you talk about frontier AI, we basically have a duopoly in this country right now, it is Anthropic and OpenAI. They have...

Well, the first thing to recognize is they have duopoly market share. You look at their growth, you look at the model capabilities, you look at the revenue, by any definition, they are a duopoly. So, when Dario and Sam come out and say, 'Well, we got to pace the frontier.' It's like, 'Guys, go ahead. You are the frontier. You can just do this.' That's what I'm saying is you don't need some special benefits from the government to do this. This is your corporate and social responsibility. So you should just go ahead and do it.

I think they're sincere about the desire to make their products safe, but I do think that when Dario says he needs an antitrust waiver, I do question that.

Well, I think we are in a political season right now and again it feels like a panic is being ginned up and I think we've lived through different panics in the past and so I do think that the threat here has been exaggerated and whipped up to the point where yeah, I think basically. And then, you know, trying to put a number on it. You know again one of the things I think is very misleading about what's happened is you have these folks who call themselves researchers, they say they work in a lab. I mean basically they're employees of a corporation but when they give these numbers like they say well this could be a greater than 10% chance of extinction. It makes it sound like they're making scientific predictions. These are not scientific predictions. These are just some people's opinion about the future, not the present but this is basically just their view of what could happen in the future. There is no evidence or facts to support that. This is their view.

Well, look, I mean, again, no one is saying that they shouldn't take steps on their own to prioritize the safety of their own products. Again, every corporation in America has an obligation...

Well, no. The hoax is that somehow we're going to be extinct as a species within a few years because of AI and therefore we have to stop completely. Look at what Bernie Sanders is proposing right now. A new bill that would basically pause all development, hand our lead in the AI race to China. There's no evidence to do something like that and it'd be foolish. So look, what I'm saying is there are risks and there are benefits. We have to manage the risks. I don't support the idea that some of these frontier lab CEOs are seem to be promulgating that it's it seems to me they're kind of trying to deflect responsibility. First and foremost, it's on them to make their products safe. They should tell us how they should do that. If they can't do that, they should step aside for people who can do that. What they should not be doing is making the public afraid that they're doing a bad job. Now look at the end of the day I think these risks can be managed and I do not want to stop. I want to proceed safely. I think some caution is warranted but we should not pause or stop. If we do the Bernie Sanders plan we will hand the AI race to China and they will benefit from this technology.

I think it depends on what layer of the stack you're talking about. I think when it comes to chips they're a couple years behind us. When it comes to models, I think most people say about six to 12 months.

Well, I would say that in Silicon Valley time it's actually pretty far. It's actually a pretty decent-sized lead but not if we do something foolish like shoot ourselves in the foot. So again, you know, we should proceed safely, cautiously. There's a balance, but if we just pause all development, they will race ahead. It's pretty clear they're not going to stop. There have actually been statements out of China, I think just today, where they basically have said, 'We're not going to stop.' They think they have a different view of this. They have a very pragmatic view of AI. They think that...

Well, I think that their approach is that they see AI as something that can tangibly benefit their population. They think it's going to help their economy. They think it helps their nation. You don't see all this hand-wringing about AI. They don't have these like doomer groups who are telling their population that they're going to go extinct in a few years and so they're plowing ahead and I think we have to realize that.

Well, I'm an American and I think that the American government should have jurisdiction and control over AI in America. It's I think we should maybe talk to other governments to see how they're thinking. Talking is never bad, but I don't really believe in this globalist approach and I don't think it's going to happen within any reasonable time frame. I mean, you want to create some sort of transnational organization to regulate AI that would take years to set up. I don't think we have years. And this is where...

Oh, it's just again I think it's going to slow us down. I mean, Dario also says that somehow...

I think we should talk to them, but I'm not holding my breath that we're going to be able to agree on a deal with China. I mean, here's the contradiction. Dario in his essay basically calls China a bunch of evil dictators. He says we should starve them for compute. We should bring their economy to its knees, not sell them chips. And then he says that, oh well, you know, we can work out a deal with them. Well, how are you going to work out a deal with all this hostile rhetoric? So, again, this just seems fatal to his plan, this idea of some sort of global governance. I think that what should happen here is that Dario should start first and foremost by minding his own lab, by basically doing a good job with the thing he can control the most, which is his lab. And Sam should do the same. They are the frontier. Those two companies are the American frontier. If they do a good job with AI safety, then you know working out a deal with China or these other countries becomes a little bit less important. And that's the thing that has to happen right now. We cannot wait years to negotiate some sort of international deal. We need Dario and Sam to basically take responsibility right now.

Well, I think transparency is good. I think auditing is good. I mean, we can work out exactly the details of how that works, but yeah, I think audits are good. I mean, I think the public should have transparency into what's going on here. And I think that these labs should be living up to their commitments. So, yeah, I think now look, I think the big question there is he calls them independent evaluators. And then we found out that the group that he mentioned is funded by his investors and has lots of former Anthropic employees there. So I just wonder is this a cozy arrangement or is this actually truly an independent group? But look, if these are independent auditors, I have no problem with that. I think that's a positive step.

Well, I think President Trump knows what to say to President Xi.

He doesn't need my advice. Look, I think talking is always good. I think you talk to your friends. I think it's even more important to talk to your adversaries.

Well, the first thing I would do is probably just have a conversation. I'd like to hear how they're thinking about AI. I mean, I hear their public statements as well. I've told you I think they see it in a very pragmatic way. They don't seem to be afraid of it, but I think it's good to talk through this issue. Do we have areas of agreement that we can potentially pursue? I'm not against having an agreement, but I'm just not placing a tremendous amount of faith that we're going to be able to negotiate something like that anytime soon.

I think it probably is. Yes.

I don't think that we're going to have the end of humanity here. I think what needs to happen is that these frontier labs need to take responsibility for their own products. I think that product safety, reliability and robustness is a core part of any product and if they do that and they basically develop good practices in their engineering in their lab I think all the problems that have happened to date they can be root-caused. We should figure out exactly why they happen, what is the diagnostic, how do you avoid them in the future and you go through that process iteratively. That's what the scientific method is about and you solve those problems. I think what's happening right now is people are extrapolating off of one event, you know, letting their...

There are alarming things about it for sure. And that's the type of thing we got to root-cause. There were also things about it that I think were frankly kind of just negligent, right? I mean, this was a sandbox testing environment that was misconfigured. So, that's how the agents got on the internet. Then the company that they broke into had credentials just on the open internet and then no one was monitoring the situation. Okay, so these are all things that absolutely can and should be fixed. So this is what I mean is look we should apply the scientific method to this. If at the end of the day, however, they don't think they can fix these problems, they don't think they can make their products safe, then I don't think they're the right people to be working on this technology and maybe they should step aside for someone who can do it safely. But I don't think they're going to come to that conclusion.