One nation, two AI futures.

Dave Bittner: Hello, everyone, and welcome to "Caveat," N2K CyberWire's privacy, surveillance, law and policy podcast. I am Dave Bittner and joining me is my co-host Ben Yelin from the University of Maryland Center for Cyber Health and Hazard Strategies. Hey, Ben.

Ben Yelin: Hello, Dave.

Dave Bittner: And, on today's show, we are once again joined by our N2K colleague and editor of the weekly Caveat Newsletter, Ethan Cook. This week, we are revisiting the hot topic of AI. Welcome, Ethan.

Ethan Cook: Hey, guys.

Dave Bittner: 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, let's jump in here, gentlemen. So we are talking about AI today as our general framework. And, Ben, before we started rolling, you mentioned that there's something kind of at the top level when it comes to this that we should really start with.

Ben Yelin: Sure. So everything else we're going to talk about kind of depends on this legal concept that we did a whole show on and it's about federal preemption. So the basic issue is that states have taken the lead in promulgating AI regulations. And Ethan, very helpfully in the newsletter, has written up a lot of the different statutes that states have passed regulating AI in various ways, so guard rails, governance, frameworks, limits on certain types of deep fakes, that sort of thing. The Trump administration has announced that they are going to be releasing an executive order intended to preempt states from enacting their own AI policies. So the idea is, I think, from their policy perspective that it doesn't make sense to have 50 separate frameworks to comply with artificial intelligence-related statutes, that it would be burdensome on the business community. They want to foster, you know, the maximum amount of innovation and outcompete our competitor countries in developing AI and it would be hard to do that if there were 50 separate state standards. Interestingly, this is getting push back not just from the political left, but also from the political right, including Ron DeSantis, which is really interesting. So, obviously, he's the governor of Florida and he believes that state legislatures, because of their police powers to protect the health, safety and welfare of their citizens, should have a role in promulgating AI regulations that fit their particular states. So the rumor is that the Trump administration is going to release an executive order that sets up a task force within the administration to initiate lawsuits against states with what they determine to be overly burdensome AI regulations. And this is kind of their third attempt at a solution to this problem. They tried to tuck a preemption measure in the One Big Beautiful Bill where the federal government, as part of that statute, would say, "All AI-related legislation coming from state governments is preempted," which you can do that, you can have express preemption in statutes. And I'll talk a little bit more about that. So they tried to do that in the One Big Beautiful Bill. That didn't work, it was taken out. Then they tried to put it in a must-pass defense policy bill, but people noticed. So --

Dave Bittner: Busted.

Ben Yelin: Yeah. Which it doesn't always happen, by the way.

Dave Bittner: Right, right.

Ben Yelin: I mean, the first time they tried to do this, no one noticed.

Ethan Cook: It almost snuck by.

Ben Yelin: And it was Marjorie Taylor Greene the day after she voted for it in the House. I was like, "Wait a minute."

Dave Bittner: Yeah.

Ethan Cook: Yeah.

Ben Yelin: "Why are we doing this AI preemption thing?" So they failed to do that twice. And so this is their third attempt. And there are several problems with this both kind of legal-wise and policy-wise. The real meat of preemption is generally that the federal government is actually putting forward its own regulatory framework. And, when they do that, it makes sense for them to preempt state laws in that area. So they've done that for a number of things. Tobacco regulation is a big one. But, of course, the difference there is they actually regulated tobacco. I mean, we have federal cigarette taxes, all different types of liability policies. We don't have that for AI. There are no existing federal AI regulatory measures or statutes. So this would be preemption in the absence of federal action, which we've never had before. The second and perhaps most problematic thing from a legal perspective is you can't preempt via executive order. You just can't do it. Now --

Dave Bittner: That was going to be my first question.

Ben Yelin: -- with the caveat like "do laws even matter anymore" --

Dave Bittner: Yeah.

Ben Yelin: -- I mean, it's like the Supreme Court can just decide that he can do this and I wouldn't be that surprised. But it's never been done in the past and, certainly, I mean, I think it would be looked upon disfavorably. The whole idea behind preemption is that Congress has the intention to occupy the field. As our first branch of government, there's some authority that Congress has usually. And, in this case, that's the authority in Article I, Section 8 to regulate interstate commerce and that they would employ that to preempt state laws. But that's like the closely considered deliberative process that takes place in Congress. It would be really hard to do that via executive order. I think they know that because they're not just saying all state laws are hereby preempted, though they may try to do that as part of this executive order. The whole task force thing is designed so that they can kind of overwhelm our legal system and overwhelm state attorneys general who try to defend their statutes in court. So you can see the process working out. State XYZ passes some type of law regulating artificial intelligence. The task force, which is going to be led by David Sacks, known Twitter poster/billionaire.

Dave Bittner: Okay.

Ben Yelin: They get their lawyers from DOJ to file suit saying that this violates their whatever federal preemption policy they put in the executive order. Even if, ultimately, these lawsuits fail, it will grind the gears of state regulation. And, if they are successful in a couple of cases, it's just going to throw the entire legal framework into flux, which I think is better for them in the sense that it might, at least temporarily, invalidate AI-related statutes at the state level. So everything we talk about here, including federal AI policy and what the states are trying to do, is largely dependent on what ends up happening with the federal preemption.

Dave Bittner: Ethan, anything to add there?

Ethan Cook: Yeah, I think Ben hit it right on -- nail on the head, right? You know, I think there is a -- it is very clear and I think the pushback from the Republican Party, especially the state Republican legislators, is obvious. You know, Ben mentioned Desantis, and I think there has been a lot of support amongst Republican federal lawmakers for a lot of these AI policies because there is no federal framework at the moment. And I think that this move is less so of along party lines and feels way more along the lines of Trump's policies, what Trump is going for, and that is a very deregulated AI system. His administration does not want a system that is overburdened at any level, federal or state, with regulations that they believe are, you know, hampering innovation or slowing innovation. If you read his -- Trump's AI Action Plan, which was released earlier this year, I believe midsummer, it highlights -- and one of the first points is accelerating AI innovation. And it directly calls out that we're trying to reduce and not pass any laws that may have any hamper on innovation in any way, shape, or form or cause things to slow down. So I think this is a direct response to that. And, very clearly, a part of this overarching goal to ensure that AI innovation stops for no one, stops for no state, stops for no federal law, etc. I think, with this, regarding the legality, I also agree that slowing things down in the courts is -- even if it's not legal, even if it -- if this order is not legal, even if this preemption effort is not legal, from a traditional standpoint, I don't think that is necessarily the goal. I think the goal is to slow this down because, while this is slowed down, while things are not being able to move efficiently, AI innovation is happening. And if that can happen, then we're still achieving the prime directive, even if it's not long term, "Oh, we've really actually set up a law." I am curious whether the Supreme Court will support, though, or not. I don't know if I would instinctively say they will because, well, yes, obviously, it's a Republican-controlled Supreme Court. They have the majority there. Obviously, this is not overwhelmingly loved by his party. It's not like this is, "Oh, this is something the entire party supports. We support killing and gutting state legis -- state supports here," etc. So I think the merit of whether this will stay or not is -- long term is somewhat up in the air for me.

Ben Yelin: Yeah, I think that's exactly right. I wouldn't prognosticate on the Supreme Court on this. I mean, just kind of reading the tea leaves of the major cases right now, it seems possible that the Supreme Court is going to go against Trump on tariffs, which means, like, they have it in them on areas where there are kind of intra-conservative divides to go against the president. Obviously, this should all be based on legal precedent and sometimes it is, to be fair. But, yeah, so I can't really handicap it except to say the federal government's here is -- argument here is very novel, that you can preempt state statutes via executive order. And you could see how that would be problematic. I mean, passing an act of Congress takes some level of consensus, some level of consideration. There are committee hearings. State officials can come and testify or they can lobby for or against that type of legislation. So, when there is federal preemption that comes out of Congress, it's at least something that's a considered process. If you are allowed to preempt state statutes via executive order, what is the limiting principle?

Dave Bittner: Right. Are we going after guns and abortion and, you know, all the hot button issues, would they just flip back and forth depending on who's in the White House?

Ben Yelin: Totally. So let's say we have a president, you know, Alexandria Ocasio-Cortez, in 2029 and she decided that, you know, the Supreme Court has said in the past that there are some -- even under the Second Amendment, there are some weapons that we can regulate at the federal level, which we have done. What if she said, "By executive order, I decide to preempt all red state laws protecting gun rights"? I do not think that would go over well. And whether --

Dave Bittner: What?

Ben Yelin: Yeah.

Dave Bittner: No.

Ben Yelin: Yeah, whether that is a too extreme example or not, I think you can see where the slippery slope goes there if this is just something you can do via executive order.

Dave Bittner: Right. So perhaps the Supreme Court would be -- or, hopefully, the Supreme Court would be forward-looking enough to try to get in front of that possibility.

Ben Yelin: Yes. One would hope so.

Dave Bittner: Yeah. Let me -- oh, go ahead, Ethan.

Ethan Cook: I was going to say, and I think this opens up a larger conversation with AI and what the administration wants to do with its action plan, which is, you know, I think with -- through its AI Action Plan and the pillars it establishes, a lot of it's vague policy goals, right? It's "we want to lead this, we want to counter China's influence." But there's not really a lot of definitive policy and a lot of it -- well, they can do some things through agency enforcement, they can do some things through, you know, interpretation of an existing law, because there is no major federal comprehensive law, there is kind of a "comma, but" of we need to also have something behind us at the federal level to support some of these initiatives. If the administration is going to take the approach of, "Oh, executive orders can just ignore things like, okay, we don't have -- we're not going to have any -- you know, any effort -- any laws that hamper innovation," well, normally, you'd have to pass some kind of federal framework that would preempt the state level. And, to Ben's point, like, there is a background at the federal level to say, "Yeah, we are preempting and here is what we're providing in lug -- in lieu of that." Now, if that's where we're going, where we're saying, "Well, no, we're not going to do that at all, we're going to ignore Congress completely and just preempt in general," I think it gets to this dynamic where that AI Action Plan, if we're going the executive order route, if we're going the executive action route at the -- within the Executive Branch to really create this policy, long term I think that changes how I view the -- that AI Action Plan and the actual milestones it's hoping to achieve.

Dave Bittner: We'll be right back. [ Music ] Let me ask you this. How much of this could be wrapped up in the notion that, at the moment, the U.S. economy is being propped up by investment in AI and that's a highly motivating factor to not let anything get in the way of AI? Is that at all a reasonable line of thinking?

Ethan Cook: I think it is.

Ben Yelin: Yeah. I mean, I think that underlies everything. It's propping up the economy right now. And, without -- some people would say it's a bubble. I am genuinely not sure. But, without the rapid growth in AI-related industries over the past year, we would have negative GDP growth and negative job growth. So it is like the key driver of our economy right now and I think, for the Trump administration, like, it's all hands on deck. We need to do whatever we can to make sure we keep innovating.

Dave Bittner: Right. So my other question is -- so let's say I am the state of California and I have these AI laws that exist, they're on the books, they are active, we're enforcing them. If the federal government files suit against California, does that in any way put California's laws on hold or in limbo?

Ben Yelin: They could get a preliminary injunction that says that California's laws are on hold pending the outcome of this case. Now, the federal government would have to show substantial likelihood of success on the merits. So that would be like a preliminary step where a judge could look at all the facts and say, "Does the federal government have a chance here to prevail?" If they don't, and if the federal government would suffer some type of irreparable -- wouldn't suffer some type of irreparable harm, then we're not going to grant that injunction, and we still will let the case go to the merits while the California law stays in effect. So whether it stays in effect or not depends on the judge applying the test we apply for preliminary injunctions, so balancing the equities, whether they're -- the plaintiff would suffer some type of irreparable harm, and, most importantly, a substantial likelihood of success on the merits.

Dave Bittner: Ethan, any thoughts on that?

Ethan Cook: Well, I can't speak on the legalese that Ben is throwing out.

Dave Bittner: That's what Ben's for.

Ethan Cook: Yeah, exactly. Just put that in your AI and it'll tell you exactly what that means. But I do think, you know, a lot of -- so several -- in the deep -- several of the laws that California is putting forward are coming into effect February, beginning of 2026. And I don't think it's a coincidence that, if this is passed, a lot of these will be attempted to be sued before they go into enforcement. You know, I think that is a huge metric because there is some major laws that California has coming out, not just in terms of very specific use cases where it's they have a chatbot companion where they're targeting, you know, and holding people -- holding companies accountable for their chatbots, encouraging harmful conversations or harmful outcomes via their -- via its features, but there is a major one coming through which is the frontier AI model, which, if people remember from last year, California attempted to pass a very similar law and failed. It did not go through. Basically, it could not gain the support. It hit Gavin Newsom's desk, Gavin Newsom vetoed it saying that the implications of this law would have far more disastrous outcomes than we would actually want despite it being passed by the state legislature. Fast forward a year, we put a new forward -- a new law forward, we change up a lot of the requirements, reduce some of them, and we now have SB 53 or the frontier AI model law, which is going to significantly put transparency requirements, whistleblower protections, etc., on frontier AI models. And these are front -- what they mean by "frontier," these are the companies that are making exorbitant amounts of money annually as well as have exceptionally high computing power. This isn't just some random small-scale AI company. We're talking, you know, the very, very largest ones being regulated by this group.

Ben Yelin: And it's not the small business who put together a little chatbot for customer --

Ethan Cook: Exactly.

Ben Yelin: -- service off hours. Yeah.

Ethan Cook: We're talking about your Anthropics, we're talking about your ChatGPTs, these kind of major models where, if they were to go wrong, if something negative were to happen, then there would be disastrous outcomes.

Dave Bittner: So, if the administration's overarching attitude here is to keep hands off when it comes to AI, where are the gaps then? Where are the places where people are saying that there's stuff that needs to be addressed that is being neglected?

Ethan Cook: For me, the biggest gap is the -- what the administration is hoping to have in terms of this laisse faire approach, this step back, "we don't want to have any -- we don't want to have any" what's the right word for it, "conflict with AI and policy." And that's very evident in terms of their goal. They want to reduce the CHIPS and Science Act's funding requirements. So, to get funding from the CHIPS and Science Act, which the Biden administration gave out lots of money for AI innovation, we want to reduce that even further so we can get more money out, we can fund more projects, we can expand the energy grid. That lack of those requirements is directly counter to what states by and large are trying to do. They are trying to raise requirements, raise transparency needs, raise reporting needs, raise incident reporting requirements, etc. I think that's a huge gap where you have one very powerful group saying, "Nope, we want to pull back on all these requirements, we want to have less federal oversight." The only oversight the Trump administration really seems to aim to -- at least in the AI Action Plan seems to care about is, first, exporting to making sure that it's only going to keep partners and allies for when we export full stack technologies. And the second is they have this weird stipulation in there about DEI and making sure that AI is not woke and something along those lines. Besides that, the Trump administration really doesn't care. They are more concerned with "how can we get it into the government as fast as possible, how can we open the door to AI innovation, how can we get the funding and the money there, how can we throw the full weight of the federal government behind these companies." And the state government's on the complete other side. That is a major gap. I think the other major gap which we have to kind of reconcile with, and this is less so the Trump administration, I think they play a role in this, but it hasn't even -- they haven't even had the opportunity to play a role, which is the complete inability for Congress to pass any sort of framework.

Dave Bittner: Right.

Ethan Cook: And that's not even Trump's fault, right?

Dave Bittner: Yeah.

Ethan Cook: Like, they can't --

Dave Bittner: Yeah.

Ethan Cook: -- we can't even blame his administration because they haven't -- the House or the Senate haven't even come to -- as a consensus as one. There's been no effort to. And I think that is another huge gap where we see states doing all this and we can complain and AI companies love to complain incessantly about "oh, my God, it's so hard to, you know, track 50 legislators." In reality, that's a little grandstanding. It's not 50. Like, there's a couple big ones and then there's a bunch of all our small, little ones that are more very case-by-case focused rather than regulating massive AI companies. But the federal government has had very little interest at the congressional level of trying to come together and make a co -- build a coalition about creating an AI framework that is meant to supersede these efforts.

Ben Yelin: Can I just throw in one interesting wrinkle from a news item that came out yesterday? And I'll just read the headline, "Trump says he will allow Nvidia to sell some AI chips in China."

Dave Bittner: Yes, we covered this on today's CyberWire.

Ben Yelin: I figured. Yeah, you wouldn't let this go without being covered on the daily. If our only interest was destroying our competitors when it comes to AI, why would, after a moratorium, would we be selling not the most valuable chips, but valuable chips to our foremost competitor? Is that not a little bit contradictory?

Dave Bittner: Well, I mean, I guess it's because money, right? I mean, Nvidia has to justify that stock price somehow.

Ben Yelin: Right. But it was the government that was stopping them from selling it.

Dave Bittner: Yeah.

Ben Yelin: Right? And now they are allowed to sell it. So, from the government's perspective -- I mean, I think, in this case, to me, it sends mixed messages.

Dave Bittner: Yeah.

Ben Yelin: If you're going to go all in on "we are going to beat the hell out of China," that's one thing.

Dave Bittner: Right.

Ben Yelin: And, like, that's their stated motivating factor behind things like preemption and the president's AI policy generally. But then this new announcement yesterday just, to me, seems to undermine that. So I thought that was something interesting. The only other thing I wanted to mention is there is some polling data -- now, this comes with a giant grain of salt because most normal people, unlike us, have never thought about this for a single second in their lives.

Dave Bittner: Okay.

Ben Yelin: But I'm going to quote it anyway. And it comes from Gallup, a poll that was taken this past May. "Which should the government prioritize when it comes to AI?" The first answer: Maintaining rules for AI safety and data security even if it means slowing AI capabilities at a slower rate, that was supported by 80% of respondents, versus B: Developing AI capabilities as quickly as possible even if it means reducing rules for AI safety and data security, 9%; don't know, 11%. So that's just context.

Dave Bittner: Yeah.

Ben Yelin: Just throwing that out there.

Dave Bittner: Well, I mean, it's -- to oversimplify things, is that kind of the fundamental thing that is at issue here is that, at the federal level, we're looking for -- or they are looking at global competitiveness and even national security when it comes to doing this as quickly as possible with no interruptions or nothing getting in the way and no speed bumps. But the states are looking out for their citizens and saying, you know, "Our personal information, our privacy, our ability to walk down the street and not have AI automatically recognize," all of those things. And those two things are at odds. And, also, just toss in there, I mean, how much of this is because our AI efforts are driven by private organizations, not -- this isn't a nuclear weapon kind of thing where it's a big government effort to develop this. Right?

Ethan Cook: I would agree. I think that the federal level is concerned with competitiveness. They are concerned, which is -- I find it very interesting because, under Biden, even taking out Biden administration, the Congress at the time was very, very concerned about China's access to chips and China's ability to get access to these chips. And one of the things that almost was bipartisan the entire time was Biden's restrictions for exporting chip technology to even including and analyze just in general. They were very -- everyone was kind of on board with "let's keep this inside the nation, control what we can." The Trump administration has had a pivot there with exporting that technology. They, on the flip side, have had some -- also some efforts to give some credit to track chips more effectively and ensure that smuggling operations are trying to crack down on. I forget what it was, but I remember seeing this earlier this year that that was a key goal of theirs was to -- they acknowledge that, regardless of whether we legally are allowing it or not, it's getting over there and the more advanced stuff is getting over there because of illegal exportation.

Dave Bittner: Yeah. And they've had some success there of cracking down on smugglers. So, yeah.

Ethan Cook: And maybe that their approach -- the administration's approach is, you know, to kind of take what you can, they're not going to stop at all, try their very best to shut down the advanced stuff going and let the weaker stuff that they can't control and, you know, just take, you know, what's the lesser of two evils, you know, and just say, "It's -- they're going to get chips regardless. Let's try and minimize the advanced chips and they kind of turned a blind eye to the lower chips." I don't think it's a coincidence that these -- this consideration comes right after, you know, Trump meets with some of these major AI companies and has them for dinner and there's a whole conversation that I'm sure gives a big insight onto how these decisions are made. I think, to your point on the state level, Dave, about security and protecting the citizens, that is a huge part. If you read some of the laws, both Utah and Colorado's AI laws are -- directly mention this. Utah has one that's regarding covered providers and what they consider are regulated industries, like the medical industry, and, if you have a license, you need -- and you're using AI, you have to disclose these requirements to people. You can't just be using AI in the medical field and not be disclosing to your citizens or your patients that you're using AI and that they're engaging with an AI chatbot, let's say, on a telehealth appointment or something along those lines.

Dave Bittner: Yeah, it makes sense. Ben, what are your thoughts on that tension between the states and the feds?

Ben Yelin: I think it's true that they have different goals. I tend to default to what our constitutional system both requires and the way it has been maintained for the past 250 years, more or less, which is that, unless this is an area of federal power, federal dominion as laid out in the Constitution or something that's inherently federal, then states, as laboratories of democracy, and I know that's a very overused cliché, but should be able to regulate things like AI when it threatens people's personal privacy or when it threatens cybersecurity or the integrity of state and local government. Now, that said, like, I think Congress is well situated, if they think that this is one of those areas where, due to the interstate aspect of it, the fact that this is something that, obviously, by definition, crosses state lines and has a huge impact on the economy, that it's better regulated at the federal level, and that's done through a considered process, I would understand that. And people representing all different types of states and districts would have an opportunity to weigh in on it. And so I am open to that as a process, but that's not what an executive order does. So that's just -- that's kind of my end all and be all here.

Dave Bittner: All right, well, before we go our separate ways here today, it is the end of the year, any thoughts for how this may play out throughout 2026? Any predictions? Ethan, what do you think?

Ethan Cook: I think that the Trump administration is going to continue to execute on his AI Action Plan through executive efforts and through executive orders and agency actions. I think the stalemate in Congress right now that has formed, and likely will get worse after midterms, is going to put a serious hamper on what that administration can do and wants to do. And I think the long-term implications of that means that there's going to be a huge push to get that done within a year. And after that year is up, it's going to largely be back and forth executive actions that really don't mean a lot. And if they can stall that process out as long as possible to execute on those goals, they will.

Dave Bittner: Ben?

Ben Yelin: Yeah, I think, as Ethan said, this could have a chilling effect on states, just the idea of being sued, even if they don't know how it would work in practice. If a state legislature is 50/50 on passing something that that might have an impact and then that would work in the Trump administration's favor. But I think there are some blue states in particular that are going to be -- like California, that are -- no matter what the Trump administration is saying are going to continue to be very active in AI regulations. I work closely with the Maryland General Assembly and I know there are going to be several pieces of AI-related legislation coming up in the next session, which starts in January. And they're looking very closely at what's going on at the federal government and what we can accomplish within our own boundaries that isn't going to violate any sort of federal preemption. So predictions about the future are hard by definition.

Dave Bittner: Yeah.

Ben Yelin: But, yeah, I kind of think that we're -- this stalemate is going to continue and we might go through 2026 without having any level of certainty on how this regulatory landscape is going to evolve.

Dave Bittner: All right, well, gents, thank you for joining us. Interesting conversation as always. [ Music ] That is "Caveat" 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. Please also fill out the survey in the show notes or send an email to caveat@n2k.com. This episode is produced by Liz Stokes. Our executive producer is Jennifer Eiben. The show is mixed by Tré Hester. Peter Kilpe is our publisher. I am Dave Bittner.

Ben Yelin: And I'm Ben Yelin.

Ethan Cook: And I'm Ethan Cook

Dave Bittner: Thanks for listening.