National Security Strategy and AI Evals on the Eve of Superintelligence with Dan Hendrycks

0:05Hi, listeners, and welcome back to KnowPriors. Today, I'm with Dan Hendricks, AI researcher and director of the Center for AI Safety. He's published papers and widely used evals such as MMLU and most recently, Humanity's Last Exam. He's also published Superintelligence Strategy alongside authors including former Google CEO Eric Schmidt and Scale founder Alex Wang. We talk about AI safety and geopolitical implications, analogies to nuclear, compute security, and the state of evals. Dan, thanks for doing this. Glad to be here. How'd you end up working on AI safety? AI was pretty clearly going to be a big deal if one would just think through its conclusion.

0:43So early on, it seemed like other people were ignoring it because it was weirder or not that pleasant to think about. It's hard to wrap your head around, but it seemed like the most important thing during this century. So I thought that that would be a good place to devote my career toward. And so that's why I started on it early on. And then since it'd be such a big deal, we'd need to make sure that we can think about it properly, channel it in a productive direction and take care of some sort of tail risks, which are generally systematically under addressed. So that's why I got into it. It's a big deal and people weren't really doing much about it at the time.

1:24And what do you think of as the center's role versus safety efforts within the large labs? Well, there aren't that many safety efforts in the labs even now. I mean, I think the labs can just focus on doing some very basic measures to refuse queries related to like help me make a virus and things like that. But I don't think labs have an extremely large role in safety overall or making this go well. They're kind of predetermined to race. They can't really choose not to unless they would no longer be a relevant company in the arena. I think they can reduce like terrorism risks or some like accidents.

2:08But beyond that, I don't think they can dramatically change the outcomes in too substantial of a way. they could, because a lot of this is geopolitically determined. If companies decide to act very differently, there's the prospect of competing with China or maybe Russia will become relevant later. And as that happens, this constrains their behavior substantially. So I've been interested in tackling AI at multiple levels. There's things companies can do to have some very basic anti-terrorism safeguards, which are pretty easy to implement. There's also the economic effects that will need to be managed well.

2:46And companies can't really change how that goes either. It's going to cause mass disruptions to labor and automate a lot of digital labor. If they, you know, tinker the design choice or add some different refusal data, it doesn't change that fact. Safety or making AI go well and the risk management is just much more of a broader problem. It's got some technical aspects, but I think that's a small part of it. I don't know that the leaders of the labs would say, like, we can do nothing about this, but maybe it's also a question of, you know, everybody also has like equity in this equation. Right.

3:22Maybe it's also a question of semantics. Like, can you describe how you think of the difference between like alignment and safety as you think about it? I'm just using safety as a sort of catch-all for like dealing with risks. There are other risks. Like if you never get really intelligent AI systems, that poses some risks in itself. There's other sorts of risks that don't run, that are not as necessarily technical, like concentration of power. So I view the distinction between alignment and safety as alignment as being a sort of subset of safety. Obviously, you want the value systems of the AIs to be in keeping with or compatible with, say, the U.S.

4:02public for U.S. AIs or for you as an individual. But that doesn't make it necessarily safe. If you have an AI that's reliably obedient or aligned to you, this doesn't make everything work totally well. China can have AIs that are totally aligned with them. The U.S. can have AIs that are totally aligned with them. You still are going to have a strategic competition between the two. This is going to – they're going to need to integrate it in their militaries. They're probably going to need to integrate it really quickly. This competition is going to force them to have a high risk tolerance in the process.

4:29So even if the AIs are doing their principles as biddings reliably, this doesn't necessarily make the overall situation perfectly fine. I think it's not just a question of reliability or whether they do what you want. There are other structural pressures that cause this to be riskier, like the geopolitics. At the highest level, like bundle of weights, increasingly capable. Like, why do we care about AI from a national security perspective? Like, what's the most practical way it matters in geopolitics or gets used as a weapon? I think that AI isn't that powerful currently in many respects. So in many ways, it's not actually that relevant for national security currently.

5:10This could well change within a year's time. I think generally I've been focused on the trajectory that it's on as opposed to saying right now it is extremely concerning. That said, there are some, for instance, for cyber, I don't think AIs are that relevant for being able to pull off a devastating cyber attack on the grid by a malicious actor currently. That said, we should look at cyber and be prepared and think about its strategic implications. There are other capabilities like virology. The AIs are getting very good at STEM, PhD level types of topics, and that includes virology. So I think that they are sort of rounding the corner on being able to provide expert level capabilities in terms of their knowledge of the literature or even helping in practical wet lab situations.

6:00So I do think on the virology aspect, they do have already national security implications, but that's only very recently with the reasoning models. But in many other respects, they're not as relevant. It's more prospective that it could well become the way in which a nation might try and dominate another nation. and the backbone for not just war, but also just economic security. The amount of chips that the U.S. has versus China might be the determiner or the determinant of which country is the most prosperous and which one falls behind. So but this is all prospective. I don't think it's just speculative.

6:41It's speculative in the same way that like NVIDIA's valuation is speculative or the valuations behind AI companies are speculative. It's something that I think a lot of people are expecting and expecting fairly soon. Yeah, it's quite hard to think about time horizons in AI. We invest in things that I think of it as like medium term speculative, but they get pulled in quite quickly. You know, just because you mentioned both cyber and bio, we're investors in companies like Culminate or Sybil on the defensive cybersecurity side or Chai and Somite on the biotech discovery side or, you know, modeling different systems in biology.

7:17that will help us with treatments. How do you think about the balance of like competition and benefits and safety? Because some of these things I think are, you know, we think they're working effectively in the near term on the positive side as well. Yeah, I mean, I don't get this big trade-off between safety and, I mean, you're just taking care of a few tail risks. For bio, which if you want to expose those capabilities, just like talk to sales, get the enterprise account. Here you can have the little refusal thing for virology. But if you just create an account a second to go and you're asking how to culture this virus and here's your picture of your petri dish and what's the next step that you should do?

7:54That, yeah, if you want to access those capabilities, you can speak to sales. That's basically in XAI's risk management framework. It's just we're not exposing those expert level capabilities to people who we don't know who they are. But if we do, then sure, have them. So I think you can, and likewise with cyber, I think you can just very easily capture the benefits while taking care of some of these pretty avoidable tail risks. But then once you have that, you've basically taken care of malicious use for the models behind your API. And that's about the best that you can do as a company. You could, you know, try and influence policy by using your voice or something.

8:31but I don't see a substantial amount that they could do. They could do some research for trying to make the models more controllable or try and make policymakers be more aware of the situation more broadly in terms of where we're going. Because I don't think policymakers have internalized what's happening at all. They still think it's like they're just selling hype and they don't actually believe or the companies that the employees don't actually believe that this stuff could, you know, we could get EGI, so to speak, in the next few years. So I don't know. I don't see like really substantial trade-offs there.

9:06I see much more. I think that the complications really come about when we're dealing with like what's the right stringency in export controls, for instance. That's complicated. If you turn the pain dial all the way up for China in export controls. And if AI chips are the currency of economic power in the future, then this increases the probability that they want to invade Taiwan. They already want to. This would give them all the more reason if like AI chips are the main thing and they're not getting any of it and they're not even getting the latest semiconductor manufacturing tools for even making cutting edge CPUs, let alone GPUs.

9:42So those are some other types of complicated problems that we have to address and think about and calibrate appropriately. But in terms of just mitigating virology stuff, just to speak to sales if you're gen and tech or a bio startup, and then you have access to those capabilities, problem solved. What is a way you actually expect that AI gets used as a weapon beyond virology and security? Yeah. I wouldn't expect a bio weapon from a state actor, from a non-state actor. That would make a lot more sense. I think cyber make sense from state actors and both non-state actors. Then there's drone applications.

10:23These could disrupt other things. These could help with other types of weapons research, like help explore exotic EMPs, could help create better types of drones, could substantially help with situational awareness so that one might know where all the nuclear submarines are. Some advancement in AI I might be able to help with that, and that could disrupt our second strike capabilities and mutual issue destruction. So those are some geopolitical implications. It could potentially bear on nuclear deterrence, and that's not even a weapon. The example of just heightened situational awareness and being able to pinpoint where hardened nuclear launches are or where nuclear submarines are is just informational, but could nonetheless be extremely disruptive and destabilized.

11:14Outside of that, the default conventional AI weapon would be drones, which is, I don't know, that makes sense that countries would compete on that. And I think that it would be a mistake if the U.S. weren't trying to do more in manufacturing drones. Yeah, I started working recently with an electronic warfare company. I think there's a massive lack of understanding of just like the basic concept of, you know, we have autonomous systems. They all have communication systems. Our missile systems have targeting and communication systems. And from a battlefield awareness and control perspective, like a lot of that thought will be won with radio and radar and related systems.

11:59Right. And so I think there's an area where AI is going to be very relevant and is already very relevant in Ukraine. Speaking about AI is assisting with like command and control. I mean, I was hearing some story about how on Wall Street, humans used to not be able to you always had a human in the loop for each decision. So at a later stage, before they removed that requirement on Wall Street, you just had rows of people just clicking the accept, accept, accept button. And we're kind of getting to a similar state in some context with AI. It wouldn't surprise me if we'd end up automating some more of that decision making.

12:38So this just turns into questions of reliability. And doing some reliability research seems useful. To return to that larger question of where are the sort of safety tradeoffs? I think it's people are largely thinking that the push for risk management is to do some sort of pausing or something like that. An issue is you need teeth behind an agreement. If you do it voluntarily, you just make yourself less powerful and you let the worst actors get ahead of you. You could say, well, we'll sign a treaty. We will not assume that the treaty will be followed. Like that would be very imprudent. You would actually need some sort of threat of force or something to back it up, some verification mechanism.

13:20But absent that, if it's entirely voluntary, then this doesn't seem like a useful thing at all. So I think people's conflation of safety, what we must do is we must voluntarily slow it down. It just doesn't make as much geopolitical sense unless you have some threat of force to back it up or some very strong verification mechanism. But absent that. As a proxy, there's clearly been very little compliance to either treaties or norms around cyber attacks and around corporate espionage, right? Yeah. I mean, corporate espionage, for instance, that was one strategy. It's a voluntary pause strategy. We believe that equals safety.

13:59And then maybe last year there was that paper, Situational Awareness, where people were written by Leopold Aschenbrunner, and he's a sort of a safety person. So his idea was let's instead try and beat China to superintelligence as much as possible. But that is some sort of weaknesses because it assumes that corporate espionage will not be a thing at all, which is very difficult to do. I mean, we have some places, 30 % plus of the employees at these top AI companies are like Chinese nationals. I mean, this is not feasible. If you're going to get rid of them, they're going to go to China and then they're probably going to beat you because they're extremely important for the U.S.'s success.

14:36So you're going to want to keep them here. But that's going to expose you to some information security types of issues. But that's just too bad. Do you have a point of view on how we should change immigration policy, if at all, given these risks? So I would, of course, claim that this is the policy on this to be totally separate from southern border policy and other and broader policy. But if we're talking about AI researchers, if they're very talented, then I think you'd want to make it easier. And I think that it's probably too difficult for many of them to stay currently. And I think that that discussion should be kept totally separate from southern border policy.

15:07Just in terms of broad strokes, like things that you think won't work, voluntary compliance and assuming that'll happen or just straight race. So we want to be competitive. And I think it's I think racing in other sorts of spheres, say drones or chips seems fine. If you're saying let's race to super intelligence to try and get and turn that into a weapon to crush them. And they're not going to do the same or they're not going to have access to it or they're not going to prevent that from happening. That seems like quite a tall claim. I mean, if if we did have a substantially better AI, they could just co-opt it.

15:43You could just steal it unless you had really, really strong information security. Like you move the AI researchers out to the desert, but then you're reducing your probability of actually beating them because a lot of your best scientists ended up going back to China. Even then, if there were signs that they were really pulling ahead and going to be able to get some powerful AI that will crush, that will enable China or that would enable the US to crush China, they would then try to deter them from doing something like that. They're not going to sit idly by and say, you know what? Yeah, go ahead.

16:15Develop your super intelligence or whatever, and then you can boss us around and we'll just accept your dictates until the end of time. So I think that there is kind of a failure of some sort of second order reasoning going on there, which is, well, how would China respond to this sort of maneuver if we're building a trillion dollar compute cluster in the desert, totally visible from space? And it's basically the only plausible read on this is that this is a bid for for dominance or a sort of monopoly on superintelligence. So I think it's it reminds me of in the nuclear era, there's a brief period where some people were saying, you know, we got to just like preemptively destroy or preventively destroy the USSR.

17:01We got to nuke them. Even people, even pacifists or people who are normally pacifists like Bertrand Russell were advocating for this. The opportunity window for that was like maybe didn't ever exist, but there was there was a prospect of it for some time. But I don't think that the opportunity window really exists here because of the complex independence and the multinational talent dependence in the United States. But I don't think you can have China be totally severed from any awareness or any ability to gain insight or imitate what we're doing here. We're clearly nowhere close to that as a real environment right now, right?

17:41No, it would take years. It would take years to do well. And I don't even think the timelines for some very powerful AI systems, there might not even be enough time to do that securitization anyway. So, OK, in reaction, you propose, along with some, you know, other esteemed authors and friends, Eric Schmidt and Alex Wang, a new deterrence regime, Mutually Assured AI Malfunction. I think that's the right name. MAME, a bit of a scary acronym and also a nod to Mutually Assured Destruction. Can you explain MAME in plain language? Let's think of what happened in nuclear strategy. Basically, a lot of states deterred each other from doing a first strike because they could then retaliate.

18:20So they had a shared vulnerability. So we're not going to do this really aggressive action of trying to make a bid to wipe you out because that will end up causing us to be damaged. And we have a somewhat similar situation later on when AI is more salient, when it is viewed as pivotal to the future of a nation. When people are on the verge of making a superintelligence more, when they can, say, automate pretty much all AI research, I think states would try to deter each other from trying to leverage that to develop it into something like a superweapon that would allow the other countries to be crushed or use those AIs to do some really rapid automated AI research and development loop that could have it bootstrap from its current levels to something that's a superweapon.

19:10intelligent, vastly more capable than any other system out there. I think that later on, it becomes so destabilizing that China just says, we're going to do something preemptive, like do a cyber attack on your data center. And the US might do that to China. And Russia, coming out of Ukraine, will reassess the situation, get situationally aware, think, oh, what's going on with the US and China? Oh my goodness, they're so ahead on AI. AI is looking like a big deal, let's say it's later in the year when, you know, a big chunk of software engineering is starting to be impacted by AI. Oh, wow, this is looking pretty relevant.

19:45Hey, if you try and use this to crush us, we will prevent that by doing a cyber attack on you. And we will keep tabs on your projects because it's pretty easy for them to do that espionage. All they need to do is do a zero day on Slack, and then they can know what DeepMind is up to in very high fidelity and OpenAI and XAI and others. So it's pretty easy for them to do espionage and sabotage. Right now, they wouldn't be threatening that because it's not at the level of severity. It's not actually that potentially destabilizing. It's still too distant, the capabilities. A lot of decision makers still aren't taking this AI stuff that seriously, relatively speaking.

20:19But I think that'll change as it gets more powerful. And then I think that this is how they would end up responding. And this makes us not wind up in a situation where we are doing something extremely destabilizing, like trying to create some weapon that enables one country to like totally wipe out the other. And as was proposed by people like Leo. What are the parallels here that you think makes sense to nuclear and don't? I think that more broadly, just as a dual use technology, dual use would be the civilian applications. It has military applications. Its economic applications are still, you know, in some ways limited.

20:56And likewise, its military applications are still limited. But I think that will keep changing rapidly. Like chemical, it was important for the economy. It had some military use, but they kind of coordinated not to go down the chemical route. And bio as well can be used as a weapon and has enormous economic applications. And likewise with nuclear too. So I think it has some of those properties for each of those technologies. Countries did eventually coordinate to make sure that it didn't wind up in the hands of rogue actors like terrorists. There have been a lot of efforts taken to make sure it doesn't, that rogue actors don't get access to it and use it against them because it's in neither of their interests.

21:42Basically, like bioweapons, for instance, and chemical weapons are a poor man's atom bomb. And this is why we have the Chemical Weapons Convention and Bioweapons Convention. That's where there's some shared interests. So they might be rivals in other senses in the way that the U.S. and the Soviet Union were rivals, but there's still coordination on that because it was incentive compatible. It doesn't benefit them in any way if terrorists have access to these sorts of things. It's just inherently destabilizing. So I think that's an opportunity for coordination. That isn't to say that they have an incentive to both pause all forms of AI development, but it may mean that they would be deterred from some particular forms of AI development, in particular ones that have a very plausible prospect of enabling one country to get a decisive edge over another and crush them.

22:34So no like super weapon type stuff, but more conventional type of warfare like drones and things like that. I expect that they'll continue to race and probably not maybe not even coordinate on anything like that. But that's just how things will go. That's just, you know, bows and arrows and nuclear. It just made sense for them to develop those sorts of weapons and threaten each other with him. If you all could propose a magical adoption tactically of some policy or action to the current administration, what is the first step here? It is the, you know, we will not build a super weapon and we're going to be watching for other people building them too.

23:08And so I've sort of been alluding to throughout this whole conversation, like what would the companies do? Like not that much. I mean, add some basic anti-terrorism safeguards, but I think this is like pretty technically easy. This is unlike refusal for other things. Refusal robustness for other things is harder. Like if you're trying to get it like crimes and torts, that's harder because it's a lot messier. It overlaps with typical everyday interaction. I think likewise here, the asks for states are not that challenging either. It's just a matter of them doing it. So one would be the CIA has a cell that's doing more espionage of other states' AI programs.

23:44So that way they have a better sense of what's going on and aren't caught by surprise. And then secondly, maybe some part of government, like let's say Cybercom, which has a lot of cyber offensive capabilities, gets some cyber attacks ready to disable other data centers in other countries if they're looking like they're doing something, running or creating a destabilizing AI project. That's it for the deterrence. For non-proliferation of AI chips to rogue actors in particular, I think there'd be some adjustments to export controls, in particular, just knowing where the AI chips are at reliably.

24:20We want to know where the AI chips are at for the same reason we want to know where our fissile material is at, for the same reason that we want Russia to know where its fissile material is at. Like it's just generally a good bit of information to collect. And that can be done with some very basic states craft of having a licensing regime. And for allies, they just notify you whenever it's being shipped to a different location and they get a license exemption on that basis. And then you have enforcement officers prioritize doing some basic inspections for AI chips and use checks in. So I think like all of these are a few texts away or a basic document away.

25:00And I think that that kind of like 80-20 is a lot of it. Of course, this is this is always a changing situation. Safety isn't as I've been trying to reinforce, not really that much of a technical problem. This is more of a complex geopolitical problem with with technical aspects. Later on, maybe we'll need to do more. Maybe we will, there might be some new risk sources that we need to take care of and adjust. But I think like right now, I think that espionage or CIA sabotage with Cybercom, building up those capabilities, buying those options, seems like that takes care of a lot of the risk. Let's talk about compute security.

25:37If we're talking about a hundred thousand networked state-of-the-art chips, you can tell where that is. How does DeepSeek and the recent releases they've had factor into your view of compute security, given expert controls have clearly led to innovation toward highly compute efficient pre-training that works on chips that China can import at what one might consider like an irrelevant scale, a much smaller scale today? It's hard for me to see directionally that training becoming less efficient, even if we, you know, people want to scale it up. And so, like, does that change your view at all? No, I think it just sort of undermines other types of strategies like this Manhattan Project type of strategy of let's move people out to the desert and do a big cluster there.