
The battle for AI regulation.
Dave Bittner: Hello everyone and welcome to "Caveat," N2K CyberWire's privacy surveillance law and policy podcast. I'm Dave Bittner and joining me is my co-host Ben Yelin from the University of Maryland Center for Cyber Health and Hazard Strategies. Hey there Ben.
Ben Yelin: Hello Dave.
Dave Bittner: And also, with us today is our N2K CyberWire colleague and editor of the "Caveat" newsletter as well as the "Weekly Signals and Space Briefing," Ethan Cook. Ethan, welcome.
Ethan Cook: Hey guys, good to be back.
Dave Bittner: On today's show, we look at the calls for an AI slowdown and the president's pushback. While this show covers legal topics and Ben is a lawyer, the views expressed do not constitute legal advice. For official legal advice on any of the topics we cover, please contact your attorney. [ Music ] All right, gentlemen, we have a lot to talk about here today, but before we do, we got a little bit of follow-up here from someone who's referring to themselves as a cyber kid from down under. Says, "Hey, 'Caveat' team, longtime listener, and was happy to hear the Australia algorithm law make 'Caveat.' Sometimes I worry 'Caveat' is heavy on the US and EU content, so well done." Fair enough.
Ben Yelin: I'm thrilled at this individual emailing us. If you know anybody in Melbourne, as now I know that's -- now I know that's how it's pronounced, who wants to host a special Australia edition of 'Caveat, it looks like a incredibly beautiful city, and I'm all in. I hope you guys are, too.
Dave Bittner: Yes, so this person says, "We have cool cyber conferences, CyberCon in October in Melbourne." Or as we say, Melbourne. It's one -- it's the worst. Have -- and he's happy that the 49ers won a game for you, Ben.
Ben Yelin: I am too. It was great.
Ethan Cook: It was great. My opponent had all their players on fantasy, and it was -- it was fantastic. The Rams did not do well.
Ben Yelin: They did not do anything. Yes, no. I was --
Ethan Cook: No.
Ben Yelin: It turns out that flying in 24 hours before is not as effective as flying in a week before and acclimating yourself to the time difference.
Ethan Cook: Jet lag is a real thing --
Ben Yelin: Yes.
Ethan Cook: -- despite some people's opinions.
Dave Bittner: Right, so did some people refer to it as the jet lag bowl.
Ben Yelin: It was the jet lag bowl, and it was a great experiment in how do you do a, you know, 16-hour flight, whatever it is. It's a 17-hour time difference, but really it's kind of a 7-hour time difference because it's just -- it's the next day and --
Dave Bittner: No, Australia is a time machine. It's already tomorrow.
Ben Yelin: The Rams looked like they were -- they were still asleep. They were still on their flat beds on Qantas Airlines, so I'll take it.
Dave Bittner: All right, well again, thank you for writing in. We do appreciate the kind words, and of course we would love to hear from any of you out there. Our email address is caveat@n2k.com. All right, let's jump into today's topic here, and we're doing a single topic episode here today, and we're coming off of the heels of a bunch of the folks who head up some of the big AI companies calling for a slowdown of AI development, as well as some pushback we've seen from the president himself, and a lot of commentary coming from all over with all kinds of opinions of what exactly might be going on here. So, why don't we start off with just a little bit of the backstory here. Ethan, do you want to start us off with what happened over the weekend? We had -- it was the head of Anthropic gave a compelling, I guess, interesting interview.
Ethan Cook: Yes, I think you know, to your point, leaders across the AI space whether that be from OpenAI, Anthropic's, etcetera, have all kind of sort of raising the alarm bell saying that maybe AI is getting a little out of control. Individually, each of them have been putting out press releases briefings not only calling for stronger regulation, but also just kind of more, let's tone it back a little bit. And I think a lot of this ties back to, and if I'm right, they -- many of them directly reference it, the model escapes that have been happening almost seemingly weekly that have been published. Now, for reference, that doesn't mean they're happening weekly currently. They over the past, you know, about month and a half, every week we hear about a new model escape that happened several months ago. And that's not limited to any one model maker. That's across the board. Any frontier model maker has pretty much been having some incidents reported. And some of these have been really concerning to mildly concerning. I think the most famous one is the Hugging Face incident. That kind of was the incident that kicked it all off. But there have been several others and OpenAI has been kind of the one I've seen who's been in the spotlight the most, but again, this is not limited to them.
Dave Bittner: Yes.
Ethan Cook: And I think the reaction has been from the broad public as well as developers. This is really concerning because we thought we had a handle on this, and it was very clear that for months we did not. And they -- and there has been real world impact on that. And I think the other kind of element there is if this is where we're at now, what does this look like in five years when we're potentially talking about fully autonomous agents and not, you know, still kind of model prediction technology?
Dave Bittner: Ethan, it's such adorable that you think we're going to be here in five years.
Ben Yelin: Yes, I was just about to say you are the -- you are the optimist among us. No, I mean, I think just to add slightly more context, and Ethan, you summarized it quite well, is there is a employee, somebody who had worked for OpenAI in the past, and he was working for Anthropic, and he resigned. I think it was a week ago today as we're recording.
Dave Bittner: Yes, his name is Jacob Coxon. Yes, he's become famous overnight. He's doing the rounds on cable news, and basically, it was a warning shot saying the frontier models are not taking this seriously enough. We're not blowing smoke up your you know what. We legitimately believe that there is perhaps a 10% chance that this is going to wipe out humanity at some point in the next decade. And there was another developer who still works for Anthropic who was like, "Yes, that checks out." What he's saying is correct. So, I mean, at least in my life, I don't know about you guys, like this raised a level of consciousness about AI safety that I have not seen yet. Yes.
Ethan Cook: For sure.
Ben Yelin: I think Hugging Face was the warning shot to people in the know, people in the community who understand what Hugging Face was, who are technologically adept to realize, like, okay, they escaped the sandbox, and they exceeded their authority, and they were communicating with one another, and we're all going to die. And I think that was understandable to people who are in the field. But this was -- this is where it broke containment to use the same term in terms of the public consciousness, and I had people in my life who I've never really talked to about this stuff being like, "What is -- what is going on?" And then I think Amodei and Altman this weekend were reacting to that reaction, and there was just this understood need to do something. I mean, there's a lot of responsibility on these frontier models, and I think the pressure was growing for them to put a pause on this.
Dave Bittner: We had an open letter from more than 1300 computer scientists that were calling for the industry to slow down and for the government to step in. And the article in the New York Times pointed out that this seemed to be a deliberate echo of Albert Einstein's famous letter to Franklin D. Roosevelt about the potential power of nuclear weapons.
Ben Yelin: Yes. Yes, it's funny how history, if it doesn't repeat itself, it rhymes.
Dave Bittner: Right.
Ben Yelin: I've seen a lot of analysis about nuclear weapons for a number of reasons. You know, the rationale of why we should not slow down is that China's not slowing down, and so it's sort of the same rationale we had during the arms race, where like yes, these nuclear weapons might destroy the world, but if we don't build them, the Russians are going to build them, and they're going to destroy us. So, there's certainly some parallels there, and yes, to me, I saw somebody say this and this resonated with me. This felt like kind of late February, early March 2020, when COVID was something that was kind of simmering beneath the surface among people who were plugged into public health, but didn't become an issue that resonated with the public where there was this type of consciousness until these high-profile things started happening. Tom Hanks got it, and they had to shut down the NBA season. I feel like that's kind of where we landed this past week.
Dave Bittner: Well, meanwhile, Ben, President Trump reacted to this in -- on social media. He posted in his own unique style, saying that it's full steam ahead.
Ethan Cook: Thank you for this attention to the matter.
Ben Yelin: Exactly. This was -- I know people who refer to his Truth Social posts as bleats because they're not tweets. He's singular in his philosophy that our only goal, our primary goal, and pretty much our only goal as it relates to AI policy is to outcompete China and to be the worldwide leader in AI development. And so, he's very critical of Amodei and Altman. I noticed he didn't mention Elon Musk who also agreed with the -- his fellow big tech overlords on this matter, but Trump was basically saying that not only are -- should we go full steam ahead, but that people are being irrational. People have this irrational fear of data centers and that's motivating this. And then, in his words, like this is just another hoax. It's like the Russia Russia Russia hoax. It's --
Dave Bittner: Well, he actually compared it to climate change, which he calls a hoax.
Ben Yelin: Which he believes is a hoax. Yes.
Dave Bittner: Right. Right.
Ben Yelin: I do not want to get overly political here, except to say that like Trump takes everything personally, and so I think in his mind, like this is a conspiracy to hurt his economy. If the frontier models start to slow down and it's AI that's propping up GDP in the stock market, that'll end up having a negative impact on him and his popularity. And so, this is the same thing that happened in COVID. He at some point decided that COVID wasn't real. It was a conspiracy to ruin the beautiful economy that he had built over his first term, and I think that colored his reaction to COVID and the policies that came from his administration. So, I don't think there's a clear argument one way or another on these matters, except to say that like, I don't think Trump is really a trusted agent on this topic.
Ethan Cook: To play not necessarily devil's advocate, because I think all of us can agree that regulation on these things is important. Back in school when I was in political -- for undergrad for political science, one of the best things I ever heard was that regardless of whoever's in power, Democrat or Republican, the State Department is conservative. And the premise of that, the, you know, kind of the argument put forward is that no administration ever wants to relinquish U.S. power. No administration ever wants to be the administration that takes a step back and says, "All right, we're going to de-escalate the U.S.'s competitive advantage in any capacity," whether that's from an international relations perspective, whether that's from a technology perspective, etcetera. And I think, given how aggressive China is with their development practices, we've talked about it before with model distillation being a thing where they are actively diluting and for lack of a better word, stealing U.S. AI capabilities, there is I think a very legitimate concern that if the U.S. does not push forward, that they will not only rapidly fall behind, but then lose their leverage point over international fields about what can and cannot be done from a technological perspective, and regulatory perspective. I think that while there obviously is -- there is a fine line between let's just go crazy and have no regulations at all, and let's overregulate, I think you know, when you look at the Biden administration with how they approached this, obviously, AI was at a very, very different point from both a capability perspective, a money perspective, a kind of entrenchment across sector perspective. They took the approach of allowing innovation, but never at the -- at the perceived consequence to the average consumer or industry. And I think that's something that we can look at here as a nice balance point. OpenAI just announced that it's backing for the UK to implement mandatory regulations for binding UK -- for a binding UK AI law. A, if the developers themselves are okay with certain restrictions, and we see that both Anthropic, OpenAI are also just as involved in the U.S. states promoting these technology -- promoting these restrictive laws, that there is a middle ground between let's go crazy with development and also let's still maintain competitive advantage, and I don't think these companies would be advocating for laws that were actively going to remove their competitive advantage.
Ben Yelin: It is interesting. I mean, it's one thing for lower-level developers to be making statements like this. You rarely see in any other fields CEOs coming to the table and saying, "Regulate us." Now, there -- among certain corners on the internet, I think the -- I don't want to say conspiracy because I do think there's some truth to this, but that this is basically regulatory capture. That if these big frontier companies are going to Congress or state legislatures and saying regulate us, they won't really be hurt by that regulation because they have such a large market share.
Ethan Cook: Right.
Ben Yelin: Like you're not going to destroy Anthropic, but --
Dave Bittner: Please don't throw us in the briar patch.
Ben Yelin: Yes, but it is going to hurt the open-source models or the smaller guys. And so, I think there's something legitimate to that, but I also think like are we really sure that this is about regulatory capture and not about these people genuinely believing that their technology might you know, condemn us to an early end to our human existence? Like I'm not sure I want to take that risk.
Dave Bittner: I guess I mean both can be true.
Ethan Cook: Yes, and to add a third lens to that I would say that there's also an aspect of they don't want to be left holding the bag when we decide that what they've been doing is illegal.
Dave Bittner: Right.
Ethan Cook: And right, like -- like --
Dave Bittner: Let me just put a point on that, Ethan, because I think it's worth reminding ourselves and everyone that things like the Hugging Face event, all of these breaches, are already illegal.
Ethan Cook: Very true, very true.
Dave Bittner: The laws are already on the books. So, we're just -- because we're in this frontier moment of pushing the boundaries, everyone is sort of holding back and saying, "Oh, how interesting! This illegal thing that you did. We're going to not go after that because we're right at the edge of what's possible." But it doesn't change the fact that it is a violation of the Computer Fraud and Abuse Act. You don't think so, Ben?
Ben Yelin: I actually have a student who's going to be writing a paper about this.
Dave Bittner: But you're illegally accessing someone else's system.
Ben Yelin: Yes, I mean I think in Hugging Face, I think it's pretty open and shut that that's a violation of CFAA. When you get into like can you impute liability on autonomous agents, I think the law is very unsettled in that. So, the Computer Fraud and Abuse Act, like there's a mens rea requirement. It requires some type of criminal intent, and depending on what theory courts follow, like can we impute that intent on an actual human being that we can punish? I'm not entirely sure. Not to mention that, like, if you look at the Van Buren case that we've talked about, there's this delineation between a -- there's like this gate up, gate down approach where if you're allowed to be on a network, but you misuse your authority, that's not going to be a violation of the CFAA. But if you like hack into a network, you know, I think it's pretty clear that it's the latter scenario when we're talking about Hugging Face. But there are going to be like other break-free scenarios where that's not necessarily the case, where like agents are somewhere where they're technically allowed to be, but they exceed their authority. They break alignment. I'm trying to -- I'm trying to adopt the lingo of our AI overlords.
Dave Bittner: There was an analogy I heard this morning that was based on real world events. If my dog attacks you and kills you, we don't sit there and go, "Oh my gosh, the dog had its own free will and what are we going to do? The dog --." Just no, it's my responsibility. My dog killed you.
Ben Yelin: You trained it and you were responsibility for -- you were responsible for instituting guardrails.
Dave Bittner: Right.
Ben Yelin: Like you are liable because you didn't put up a fence. You should have known that your dog was dangerous.
Dave Bittner: Right. So.
Ben Yelin: The thing is though like our legal system has not --
Ethan Cook: There's no laws that update that right.
Ben Yelin: As of right now, like our legal system has not universally adopted that framework, where it's something like products liability where you don't have to prove negligence. You just have to show that these people created the product, and whatever happens downstream from that, they are liable for. Maybe that's where we get eventually, but like right now it's very unsettled.
Dave Bittner: But I feel like again, I've heard folks talking about how we need to change the way that we're talking about this, and that we need to stop saying AI could kill us all, and say the companies who are building these AI models could kill us all.
Ethan Cook: Yes, I think there's a dynamic of like -- and this isn't like a niche take that modern law is not equipped to handle any of these technologies and it is vastly outdated and you can you know, theorize your way out of it by saying, "Well technically, I didn't do it. It was the agent that did it," right, whether or not you own it or not. And I think there's a kind of this broad sentiment and I've seen AI developers push for this as well saying that we need to update laws to actually hold -- to clear this up, and I think to again to their point, it's for them as well, so they understand exactly what is legal and not legal, and so they can make sure that they sidestep any issues and/or not get caught five years later going, "Oh well, now we have a massive legal case on our hand." The concern that I have seen with updating a lot of laws or putting new laws is that by the time they get drafted, debated, commented, updated, passed, signed into law, they're already out of date.
Ben Yelin: Yes. We're going to be dead. We're going to be dead by that point.
Ethan Cook: Like the amount of time it takes to get there is so astronomically long compared to the amount of progress that we're seeing in these models. And there has been some discourse that I've seen about instead of trying to make hyper specific laws that target specific models or specific capabilities, you have -- you would invest the power into an agency, at least within the U.S., similar to kind of, you know, we trust the FCC to handle these things and we kind of just let them do their thing, and that gives us the power to have a far more flexible and adaptable enforcement regime. I think the caveat, you know, for the lack of a better word, is you then have to have an administration that wants to go after that and actually enforce these things, which obviously the Trump administration has no interest in doing. So, I think that's why we see a lot of the major AI companies going at the state level, because that's the only real way as the federal government doesn't have a really vested interest in making a comprehensive framework. I also think to push back, not push back, but I to provide another, you know, outside or another fold to the illegal aspect to it. I think there's -- it's not just about what their agents are doing. I think it's how they create these agents. How are they using data? How are they ingesting data? Right? What copyright laws, etcetera? And I think that there's a huge legal aspect there that we do talk about, but also doesn't get talked about nearly enough, where, you know, if you publish anything, it's being scraped within like a day and fed into an engine that they're using then to make profit. And how that data is handled, especially in terms of sensitive information, is very, very vague and not defined at all.
Ben Yelin: Yes, it's a black box.
Dave Bittner: True.
Ben Yelin: Yes.
Dave Bittner: The former chair of the FTC weighed in on elements of this and Ben I'm going to give you the opportunity to do some legal nerding out here. This is Lina Khan who again former FTC chair, said -- actually pointed to a 1934 U.S. Supreme Court decision.
Ethan Cook: Talk about legal nerdom.
Ben Yelin: Yes.
Dave Bittner: This is FTC versus R.F. Keppel and Brothers. Includes a passage where the justices argue that if keeping up with the competition requires companies to quote, "descend to a practice which they are under a powerful moral compulsion not to adopt, that competition is unfair whether or not it's criminal."
Ben Yelin: So, I think Lina Khan is somebody who's very popular on the left end of the political spectrum. So, I take what she says very seriously. I think this is an interesting argument. For one, I don't know if that case would survive the current Supreme Court. Just my personal opinion. Like, they've overturned a lot of decisions over the past several years, including some from the 1930s. The one about Humphrey's Executor, which is about restricting presidential appointments, where you can only fire somebody for cause. Like that was overturned, and I think this might be in the same vein because we have a Supreme Court that is pretty economically conservative, at least. So, that's one aspect of it. I also think that like Lina Khan's theory here requires an FTC that actively believes that these models present some type of existential risk. These are discretionary enforcement powers. So, the FTC, just like any agency, has bandwidth to enforce laws a certain way or not, depending on the priorities of the administration. So, like I don't think you're going to be successful with the legal case trying to compel the FTC to issue regulations to stop these frontier models. I think it's going to have to be a priority of the agency itself, and I think this is going to be a big issue in the next presidential campaign. It's like, are we going to use the tools of government to try to attack this problem before it goes beyond us? You know, the problem is that that's more than two years away, so a lot of bad things can happen between now and then. I mean, I think like I'm unclear on when the doom scenario is going to be fully upon us. Like I think --
Dave Bittner: Well, nobody knows.
Ben Yelin: Nobody knows. I think you could make the argument --
Dave Bittner: If ever.
Ben Yelin: Right. And I think you could make the argument that it's upon us already if you look at something like Hugging Face and then you know, depending on who you talk to, it's anywhere from one to ten years away from now, or non-existent at all if you talk to certain people. So, yes, I mean I do think it's an interesting theory. It's interesting as a legal argument. I don't think it's going to have major currency in the short term.
Dave Bittner: Yes, gentlemen, there is a lot more for us to discuss here, but first we got to pay some bills. Let's take a break here to hear from our show sponsors. [ Music ] And we are back. Let's talk about this notion that the AI companies want to be regulated. And I want to start off here just by, I don't know, pointing out the comparison. We compare this to the development of say, nuclear weapons. I think it's an important distinction that nuclear weapons were developed by the government.
Ethan Cook: Yes.
Ben Yelin: Right.
Dave Bittner: These AI models are not. In this conversation, there are folks who are saying, "Should we nationalize these models?" Where do you guys fall on that?
Ben Yelin: I declare Defense Production Act. So, I think there's a theoretical case for it. The thing is, like, if you take over these frontier models, the barriers to entry are significant, but they're not absolute. So, like, other companies could spring up that eventually could develop the same level of computing power. And is the government just going to come in and take over those companies as well? The broader question, though, is like, do we actually want that? Do we want the government to take over these companies? Because at this point, if you are pro-regulation, I think we are much better off having the companies self-regulate than subjecting themselves to government regulation at this point, considering who's in charge of the federal government and who has the most influence in AI policy. And that's David Sacks, who even this weekend with all of this is like, these frontier models are full of it. Like, it's full speed ahead. They're overstating the risks. This is about regulatory capture. Blah blah blah. So, I'm not sure if even if your first priority was safety, if right now is the proper time for a government takeover.
Dave Bittner: What about the notion that part of what's going on here is that the CEOs are trying to buy time? They see the writing on the wall that there is no direct path to profitability. Money is just being shoveled their way. That there's an argument to be made that they will never make the money back. So, perhaps a way out of this is to have a government bailout, to have the government take a large interest in this, and then we don't have to worry so much about profitability.
Ethan Cook: I would be more inclined to take that approach than the former. I, you know, I think there's an aspect of -- how much money they lose is mind-boggling, and despite that, the amount of investment that they go into, it always goes up and up and up, despite them continuing to lose crazy amounts of money. I don't know if there's a for each individual or every you know, major one. I can say that that's the case that there's like this you know, for lack of a better word, conspiracy or like agenda that that's what they want to push forward, but I do think there's a nice off-ramp for them, right? There's -- AI is proven to be one of the only things that spurns economic growth right now within the U.S., and that gives these companies, especially the major ones, a massive amount of leverage over the government, over financial sectors, etcetera.
Dave Bittner: Right.
Ethan Cook: And I think there is a kind of aspect where they're both holding all the chips. And also, at the same time, you know, kind of have the ability to trip and fall at any given moment. And I think there's a -- I feel like there is this -- I don't really have a thing to point to, but just this kind of feeling in my gut that there is a -- they're waiting for the opportune moment to sell a massive amount of IP or technological capabilities to the government or/and similar to like the Intel case where the government took a massive stake in the company. Having something along like that happen, where maybe they don't sell all of it off, but they sell a substantial portion off, offload a ton of the risk. No one cares if they're profitable anymore because it's a government service now. So, no -- the profitability within the government is a made up a term. So --
Dave Bittner: Right, well, but I mean, you think about the, let's call it an arms race with China when it comes to this. And I'm reminded again, looking back in history of our journey to the moon, you know, our journey to the moon certainly it was part -- partly about American innovation and the expanding exploration to the moon, but a big part of it was getting there first, right, before the Russians. And that's why we threw so much money at that was to beat the Russians to the moon. And so, I feel as though we're echoing that with our desire to stay ahead of the Chinese, which is a legitimate national interest. Ben?
Ben Yelin: I mean, if we're going to -- if humanity is going to be killed off, we need to do it the American way, not the Chinese way. I saw somebody brought this up. I think this is a really interesting point. Like, yes, I think the risk of being outcompeted by China, where obviously you're talking about an authoritarian government, it's going to be a very different regulatory scheme. But would the authoritarian Chinese government really be okay with out-of-control models where agents are hacking into networks? Like, don't you think they would put some controls on that for their own purposes? And therefore, like, shouldn't that impact our view of how far China is willing to go?
Dave Bittner: Well, wouldn't you think it's time for some global treaties? Shouldn't --
Ben Yelin: Yes.
Dave Bittner: -- I hate to -- it pains me to politicize this because what I really want to say is any other administration would be -- would have this underway, and I'm not sure if that's a fair statement but like Ethan referred to, my gut says it is.
Ben Yelin: You know, I think the Trump administration and I'm going to put my MAGA hat on here. He has a pretty good relationship with President Xi. They have a summit set up at the end of the month. I know there's been speculation that they might discuss AI. I don't think it's been confirmed one way or another. Maybe this is an opportunity for him to engage in diplomacy. And he's wanted a Nobel Peace Prize for a long time. This is your shot, Mr. President. If you can come up with some type of multilateral agreement that China --
Ethan Cook: SALT treaty for AI.
Ben Yelin: Exactly, and like the SALT treaty worked. Like we are much safer in this world from the threat of nuclear weapons because of the SALT treaty and other treaties like it.
Dave Bittner: Call it -- we'd have to call it the Trump treaty though, right?
Ben Yelin: We would, yes.
Dave Bittner: Well, if it saves the world, we'll do it.
Ben Yelin: I'm all for it.
Dave Bittner: Right.
Ben Yelin: But if he's able to pull this off, I'm fine with Trump treaty.
Dave Bittner: Right.
Ben Yelin: I'm fine with you can put his name on the Trump Kennedy Center.
Ethan Cook: That's if it doesn't go bankrupt.
Ben Yelin: Yes.
Ethan Cook: There's a lining with like the competition with China, and I think in terms of whether or not they would enforce or not or control, and I -- part of me says like maybe if you go that route, but I also think authoritarian regimes don't necessarily care as long as the impact is not perceived on them and is perceived on an opponent, right, or a hostile nation, in this case the US or Western society for a broader scope. I think with China and with you know, this off-ramp, I think both sides are, similar to the SALT treaty, are only going to offload if both sides can be held accountable to that offload. A non-binding treaty is all cute and all but whether or not --
Ben Yelin: Well said.
Ethan Cook: -- who cares kind of thing, it's like that's all sweet. But to me, like the value of SALT was it had instant you know, impact on removing nuclear weapons. And we said you know, we could already blow the world up 5,000 times over. Do we need to -- what's the difference between 1,000 and 5,000? Might as well de-escalate, right? And the -- and sure, we can still blow up the world a thousand times over, but I think the aspect of that's 4,000 or X amount you know, thousand numbers of nukes that we don't have to monitor, make sure are being controlled properly, access is being you know, maintained. Just facilities aren't, you know, openly creating tensions like, let's say, with the Cuban Missile Crisis, where it's like, oh, you're putting nukes right on our doorstep. I think the same kind of premise will matter with AI, where it's not about that the capabilities won't be there. They're always going to be there, and especially with these model escapes. I think it's very clear that the capabilities are closer than people feel like they may have been like eight months ago. It's more so ensuring that the control of those capabilities are far more increased, less diluted, and the understanding of those impacts are very, very clear.
Dave Bittner: Yes, will there be an AI Chernobyl? Will there be an AI Three Mile Island? That's what we're trying to prevent.
Ben Yelin: In the words of former secretary, was it Condoleezza Rice who says this or Colin Powell? We hope the smoking gun doesn't come in the form of a mushroom cloud.
Dave Bittner: There you go.
Ben Yelin: That metaphorically for AI.
Dave Bittner: Yes. So, what might regulation look like? I've heard some people say that, you know, there should be some sort of oversight, someone who has authority over these companies and has subpoena power.
Ben Yelin: Yes. So really, there are a number of approaches you could take. There's a really interesting New York Times op-ed that went through some of these approaches. I mean, one of them is to do what Bernie Sanders wants to do, and I'll say it's not limited to Bernie Sanders. Like there are people on the ideological right who support this as well, which is a -- basically, a shut it all down approach, a ban on superintelligence and a moratorium on further AI development generally. This is something that's been proposed in Congress. Obviously, this would have the biggest impact on the economy. The second approach, which I call the Dave Bittner approach --
Dave Bittner: Oh my.
Ben Yelin: -- is to treat this like we treat transportation accidents. So, aircraft crashes, we have the National Transportation Safety Board that goes out. They're dispatched to the site. They gather evidence. They conduct interviews. They figure out exactly what happened and how we can prevent that in the future. That's something that I know you've suggested for cyber-attacks generally for a long time.
Dave Bittner: Right. An FAA for these things.
Ben Yelin: Yes, I mean, I think that's reasonable. One thing you saw with Hugging Face is OpenAI released its own report, but it was -- is it Meter, M E T R?
Dave Bittner: Yes, Meter, M E T R.
Ben Yelin: Yes, that did the third-party report that I think adds more credibility. So, if you have like a non-biased, an unbiased entity come in and do a report like that, that could be a different approach. And then, you know, there's giving government auditors more access into these systems. And this would kind of be like a permanent monitoring approach, where all AI training runs are tracked in a publicly available database. Researchers are required by law to post public information on who's conducting the training run, which data center is doing the training, etcetera, etcetera, etcetera. And then the last option that's mentioned in this article that also has some bipartisan support is the so-called kill switch approach where -- and one of the bills that has a kill switch is -- that's been proposed in Congress would give DHS, Department of Homeland Security, the authority to shut down a dangerous AI operating beyond its parameters. I'm not sure that would work. Like, I don't think so. Practically, yes. I mean, for one, it's like, okay, they're trying to shut down our critical infrastructure. Like, our critical infrastructure might rely on these systems, so like, you don't want to throw out the baby with the bathwater. And I think the agents, if they're smart enough to break alignment and try to kill us all, and the models become recursive, that they're learning more over time, they might outsmart our ability to unplug them.
Dave Bittner: Right.
Ethan Cook: Yes. That was my thought when I read that law. I was like, wow, that sounds really sweet in practice and absolutely no teeth in actuality.
Ben Yelin: Yes, and I think really the upshot of all of this, all of these laws is like we should probably try everything. If we really think the problem is as severe as the CEOs are telling us that it is, then we should do all of the above. I think that was really the thrust of Amodei' letter. It was like, look, I would love international regulation, some type of treaty between us and China. If that doesn't happen, here are some ideas for domestic regulation. If that doesn't happen, here are the voluntary steps that we're going to take, and we encourage Sam Altman and OpenAI to do the same. We encourage Elon and Grok AI to do the same. So, just as many layers as are practical are the ones that should be adopted.
Dave Bittner: Well, I mean, so let's look at the practical realities here. We've got midterms coming up. After that, we're a couple years away from a presidential election. What do we suppose the possibilities of seeing anything substantive make its way through? Let's start at the federal level. Any hope for something like that?
Ben Yelin: No.
Ethan Cook: Nope.
Ben Yelin: At least in the short term.
Ethan Cook: I would be shocked if we get anything through at the federal level. That's like actual substantive and like matters. I think states are where we're really going to start seeing some of this. We already have -- California passed. It's SB 53, which is the Transparency and Frontal -- Frontier Artificial Intelligence Act in 2025. And OpenAI has pointed to this as a model for what it would like to see as like across the board over this past week, across states in some various shape or form. Obviously, it's going to be slightly different from state to state, but the law imposes not only transparency requirements, critical incident reporting, whistleblower protections, civil penalties and public safety frameworks for frontier -- exclusively for frontier models that make I think 500 million or so annually in gross revenue. Not gross profit. Gross revenue.
Ben Yelin: Profit? What's that?
Ethan Cook: Exactly. And I think that each of these you know -- and obviously I think you can say that the SB 53 in this context, it was passed before everything has kind of emerged. Needs to be updated and probably is not as teethy as we would like it to be, but I think it's better than nothing. And I think the -- within the U.S. over the next two years, if we can get a 40 out of 50 states establishing a similar federal or similar, you know, kind of AI regulatory regime, that makes passing it on the state level so -- or on the federal level so much easier. As everyone kind of already has agreed to it and it just kind of goes through and simplifies. Obviously, two years is a long time, but I just don't see at the federal level the urgency happening with the administration in its approach to actually -- unless if the states are leading the way.
Ben Yelin: Yes, I think that's 100% right on. I think state legislation, there are limits to it, but can also do a lot. We've seen that in the world of data privacy. Yes, CCPA only applied to California. It didn't necessarily on its face apply extraterritorially, but because of California's market share, like it really changed everything as it relates to data privacy in the United States. I think the same sort of thing could apply to the types of laws that California is considering passing right now. Many of them are on Governor Newsom's desk. Newsom himself has been, I think, cautious about regulating AI over the past few years. And I'm wondering with his presidential ambitions, given what's gone on the past couple of weeks, is if that's going to start to change. For example, there was a kill switch bill that came up in California that he vetoed relatively recently, and I wonder, just given the public outcry here, whether that would have come out differently if it happened now. So, I agree with Ethan. I think we can do a lot at the state level. It's not going to solve all of our problems, but it's probably the best hope for people who want regulation.
Dave Bittner: Yes, I think you can't dismiss the -- how much the public is against a lot of this stuff, how much resentment there is. You know, we see this in the fights against data centers. People are afraid for their jobs. You know, they like to play with these tools, but ultimately, everybody -- I think there's this feeling of foreboding, and I think people can make political hay out of that.
Ben Yelin: Absolutely, I think there is a lot of skepticism out there, and there's going to be, I think, a big public outcry if we do have an AI Chernobyl, where it's -- if we survive. The fact that we've even gotten to this point, I think it's going to make people very angry and justifiably so.
Ethan Cook: Yes, I think the how -- I think the -- I mean historically, this isn't even me being like oh whether I want one side to win or not. Historically, midterms always swing against the party in power, so it's likely based on just historical evidence that the Democrats take the House, potentially the Senate, based off historical data. Given the overwhelming, I think, negative view of the Trump administration, I wouldn't be surprised if it's not just a little -- by a little margin. It's a lot of margin, a lot of a margin. I think depending on how wide that margin is, you could see a world where if the coalition is big enough, you can get around the presidential veto. And I think that is not -- I wouldn't put it at a high likelihood, but I think that's something that's on the table depending on how midterms go for either party. And if that is on the table, I think that dramatically changes what can be done at the federal level. Yes, there are several Republican both representatives and senators who are very, very much in line with regulating AI --
Dave Bittner: Yes.
Ethan Cook: -- against the Trump administration. They would be, I think, very easy to convince to override a veto depending on the legislation and how it's worded.
Dave Bittner: Yes. All right. Well, gentlemen, we're going to have to leave it there. Time will tell, of course, as it always does. But I appreciate the thoughtful conversation. And again, to our listeners, if you have any thoughts or comments or would like to suggest something for us to discuss, please email us. It's caveat@n2k.com. [ Music ] And that is our show brought to you by N2K CyberWire. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a rating and review in your favorite podcast app. This episode is produced by Liz Stokes. Our executive producer is Jennifer Eiben. The show is mixed by Tre Hester. Peter Kilpe is our publisher. I'm Dave Bittner.
Ben Yelin: I'm Ben Yelin.
Ethan Cook: And I'm Ethan Cook.
Dave Bittner: Thanks for listening.

