Katherine Forrest: Right, exactly. And then the third piece of this December 11th executive order that we're going to talk about is a requirement that the Federal Communications Commission, the FCC, adopt a reporting and disclosure standard for AI models that would—and the word preemption here is actually used—that would preempt conflicting state laws, plus the Federal Trade Commission, the FTC, has to issue a policy statement laying out how the FTC Act’s prohibition on deceptive acts and practices can be used to apply to AI models. That's sort of, for those who are following at home, that's typically considered to be Section 5. And, in particular, that policy statement is going to also have to detail the circumstances under which laws requiring changes to the, you know, quote, truthful outputs of AI models—or in quotes, “truthful outputs”—are preempted by that prohibition. And then, finally, the White House has to prepare a recommendation for Congress on a legislative framework that would preempt—actually again using the word preemption—burdensome state AI laws. So with that background, let's talk about some of these legal bases, and let's start with preemption, because preemption is—it's really a term of art, and sometimes people talk about it loosely, and it's actually used a couple of times in this executive order. So let's sort of dig down into that.
Scott Caravello: Yeah, great idea. So preemption, as a term of art, stems from the Supremacy Clause in Article VI of the Constitution. And so that establishes that the Constitution, federal laws, and treaties are the supreme law of the land. And so that's the rule of decision by which conflicts between state and federal laws are decided, with the federal laws coming out on top. And so at the Constitutional Convention, going way back, there were competing visions for how federal law should actually displace state law in the event of a conflict. And so what was known as the Virginia Plan would have given Congress a direct veto over state laws, but instead the Supremacy Clause that we mentioned, which was part of the New Jersey Plan, carried the day and ensured that when Congress legislates within its authority, federal law automatically wins out.
Katherine Forrest: And I'm sure that you remember the Virginia Plan and the New Jersey Plan—that those were like just at the tip of your tongue when you were thinking about this.
Scott Caravello: That's the last thing I think about when I go to bed and the first thing I think about when I wake up in the morning.
Katherine Forrest: Right, exactly, exactly. So what the courts have done—let me sort of just move on here away from the Virginia Plan and the New Jersey Plan. So the courts have broken down preemption into three categories, and we're going to talk about those. So we have this thing called express preemption, and that's when Congress—so you have to have Congress do it—passes a law that explicitly displaces state laws on a particular subject. And the one that most people know about is, you know, they think about the U.S. Copyright Act or the Patent Act, and those two statutes preempt any state laws that attempt to regulate the subject matter of copyrights and how copyrights occur, and patent law. You don't have state patent laws. You don't have state copyright laws. So you'll sometimes see explicit statutory language that bars state laws related to a covered topic and that forbids any state requirements that are in addition to or different than the federal obligations, or that allow states to regulate only until federal rules cover the field. So you can have sort of, you know, preemption that will come into effect once, you know, states can actually regulate until that federal rule comes into effect. But these formulations have different scopes and all are explicit signals of congressional intent to override state law. And then there's something that's a second kind of preemption called “field preemption.” And I have to say I always think of field theory and Einstein, but that's probably not the last thing that you think about when you go to sleep at night. Anyway, it's called field preemption. And this is where, even if Congress doesn't use those magic words rising to the level of express preemption, Congress has legislated on the topic to such a degree, and it's so pervasive, or the federal interest is dominant enough, that there's no room for state regulation. And so immigration is a classic example of this. And then the third thing, that kind of preemption, that we're going to talk about is called “conflict preemption,” which is where state and federal law clash, but it comes in different flavors. The first type is called “impossibility preemption,” when there's simply no possibility to comply with both federal and state laws—they're just inconsistent. And the second is “obstacle preemption,” where a state law frustrates the objectives of a federal law, and this one requires courts to carefully analyze the federal law's purpose and structure. So… you take it from there!
Scott Caravello: Right. Sure, sure. And so, you know, there aren't any federal AI laws in place that would expressly preempt state-level AI laws. And so, for that reason, states have stepped into the void. And so there's a few examples, a number of which we've talked about on the podcast. California has enacted a slate of AI laws—most notably there's SB 53, which sets up a transparency framework for frontier models. Then there's Colorado's SB 24-205 that covers algorithmic discrimination and imposes governance and risk management requirements on developers and deployers of high‑risk AI systems. And then there's also Texas's law referred to as TRAGA, which stands for the Texas Responsible AI Governance Act, which does a number of things, but importantly, it prohibits certain AI uses like manipulation.
Katherine Forrest: Right. And then we've also got a whole host of algorithmic bias laws across the country, the RAISE Act. So basically we've got this panoply of state laws that have come into effect, some of which are incredibly broad. And now, December 11th, 2025, in comes the executive order and the legislative recommendation that is part of that, and the FCC reporting and disclosure standard, and the FTC, you know, policy statement—all of which are in this relatively short executive order. And so we end up having a requirement by the federal government for a legislative framework that's pretty straightforward. But let's pause on the FCC, Federal Communications Commission, because that piece is really interesting.
Scott Caravello: Totally agree. And so I think the first thing to point out about that is that the scope of the preemption is a bit unclear on its terms, right? They have to set out a disclosure and reporting standards that would preempt conflicting state laws. So presumably it would be meant to cover state-level transparency obligations like California's SB 53 that I just mentioned before, but exactly how far-reaching it would go is not clear from the face of the order.
Katherine Forrest: Right, and the standard is going to have to be grounded in the FCC's existing authority because an agency, you know, can only regulate, including to preempt state law, when it's acting within its congressionally delegated authority. So courts have repeatedly said that when an agency actually lacks the authority to regulate, it necessarily lacks the authority to preempt. So even though the executive order directs the FCC to take actions relating to these disclosure standards, those standards for AI models still have to have a nexus to the FCC’s Act, the FCC's authority, or another act of Congress.
Scott Caravello: Yeah, and so then I think that you can contrast that pretty directly with the FTC piece, which makes that nexus clear in the text of the executive order, right? The theory being that the FTC Act prohibits deceptive acts or practices. And so if there are laws that would, in the federal government's view, mandate deceptive conduct in AI models, that would be preempted. And so I think a theory here, for example, would be the Colorado AI Act—that SB 24-205—that requires reasonable care against algorithmic discrimination. And so if, in effect, you have to alter the content of the outputs to comply with that law, if you are changing the truthful output and therefore engaging in deception, you would be conflicting with the prohibition in federal law. But at the end of the day, it would be up to the courts to interpret whether there's a preemptive effect.
Katherine Forrest: Yeah, and let's just sort of pause on that because we're not saying that determining whether or not there's algorithmic bias and mitigating for that is deceptive. What we're doing is we're sort of playing out one of the ways that this executive order might be intended to be interpreted. We don't know yet, nobody's interpreted it yet and we don't actually have any guidance, but what we're looking at is how can these mandates actually sort of run smack into some of these state laws. So moving on to the litigation task force, and I knew I would have to talk about the Dormant Commerce Clause, but I'm now pushed into the Dormant Commerce Clause because to talk about the litigation task force, I have to do it. Okay, but I did learn about the Dormant Commerce Clause. I just learned about it when I was a judge. It did take me that long. All right. So the Dormant Commerce Clause is one of the key grounds on which the task force is instructed to consider challenging various AI laws that are at the state level. And these domarnt—dormant—sorry—these dormant, you know, I can't even say it five times fast. These Dormant Commerce Clause concerns emerge really given the nature of AI and its reach across state lines. And its background: under the Interstate Commerce Clause, Congress has the right to regulate interstate commerce. That's sort of like one of the basic things that it does. And that gives rise to the implied principle—and it's this implied principle that is referred to as the Dormant Commerce Clause—that states, in turn, cannot legislate to discriminate against or excessively burden interstate commerce because that's Congress's prerogative. And basically it says the states can't overly interfere with interstate commerce.
Scott Caravello: Yeah, and so it seems like, if you're reading the text of the state laws, that they've tried to scope their regulations in a way that avoids that exact issue. And so, again, going back to Colorado's law, it's limited to AI systems that impact Colorado residents. California's SB 53 appears to apply to models made publicly available to Californians. And then Texas's law has a number of hooks, but they're all tethered to the state, like when an AI system is developed or deployed in Texas.
Katherine Forrest: Right, but models, systems, and their outputs are absolutely capable of crossing state lines. I mean, you can't really cabin this kind of distributed technology. And so it's not exactly practical or realistic to say that a business needs to develop or deploy different models or systems for each state. I mean, that's not going to happen. They couldn't do it. It wouldn't be effective. They wouldn't be useful. And so the concern is that, in effect, these state laws are actually imposing nationwide requirements because, definitionally, the models are extending, or the deployment is extending, or the output is extending beyond the states, therefore—these state laws, therefore—violate the Dormant Commerce Clause because they're unduly or overly interfering with interstate commerce.
Scott Caravello: Right, but I–I think then it's helpful, Katherine, just to say that it's also at the same time not so straightforward as to say because the state laws are likely to have an impact on interstate commerce, they're likely to be struck down.
Katherine Forrest: Right. I mean, it's really difficult to predict right now what the outcome of any challenge is going to be, what the nature of the challenges will be, how the courts are going to react. You know, it's going to be up to the courts ultimately to interpret all of this. And these are the kinds of issues that could well find their way up to the Supreme Court. We'll see. And there's a lot of road between here and decisions on what actions this litigation task force might actually take up.
Scott Caravello: Well, I think that brings us to the last lever we need to discuss.
Katherine Forrest: Money?
Scott Caravello: Money.
Katherine Forrest: What is that? Is that from Cabaret—“Money, Money, Money”?
Scott Caravello: If you say so.
Katherine Forrest: What's the—yeah. Yeah, there is. Yeah, no, there are some people in our audience—I'm not going to sing it. I'm not going to burden anybody with breaking into song. But I believe that there's like in the Kit Kat Club, there's that song “Money, Money.” Anyway, let's talk about money.
Scott Caravello: All right.
Katherine Forrest: And so the big, beautiful bill, with its proposed moratorium, it was going to tie state access to a pool of funding to whether or not, in fact, the certain state laws burdened the pursuit of AI dominance. And here again, we see money—as you call—a lever. That must be your accent, wherever it comes from. Where's your accent come from?
Scott Caravello: Uh, you can't tell that that's, uh, clearly a Long Island term?
Katherine Forrest: Okay, okay. I would call it a lever, but you call it a “lee-ver”—tomato, whatever. And so, basically, what the executive order does is it says that there can here, again, be a tie of state access to some funds that are going to be federally distributed to whether or not they have onerous regulations.
Scott Caravello: Right, and so that is not a preemption issue, just to draw those lines.
Katherine Forrest: Right, it's not a preemption issue. It's a way of—let's just call it again like the stick, the carrot and the stick—it's a way of motivating states to actually comply.
Scott Caravello: Yeah, and so, looking to the authority of Congress to actually do this sort of thing, the draft provision in the big, beautiful bill that we mentioned—there's a classic case about this type of arrangement, and that's called South Dakota v. Dole, which concerned a federal law that directed the withholding of federal highway funds from states unless they set their drinking age to 21. But a complicating issue in that case was that Congress didn't itself have the authority to set a nationwide drinking age. And while we're on that, Catherine, I think–I think we can probably say that this is the first time the 21st Amendment and the end of Prohibition has come up on this podcast.
Katherine Forrest: Where–where is the 21st Amendment?
Scott Caravello: Oh, so the 21st Amendment gives the states the ability to regulate alcohol. And so Congress didn't have the authority to do so. Sorry, I should have made that clear.
Katherine Forrest: Oh! Okay, God, this is like you really took a great—you took a great Con Law class! I—and by the way, I never ran into Prohibition when I was a judge. It was because it had been repealed. Okay. I did, however, run into the South Dakota v. Dolebecause I am of an age where I was grandfathered in—I was allowed to drink at the age of 18. They passed the law, but they grandfathered in everybody who was 18. And because of this highway funding, basically every state adopted it. Today, nobody is really debating that Congress has the authority to regulate the Interstate Commerce Clause and to do what South Dakota v. Dolewas talking about. But in Dole, the Court found that the conditional spending scheme was a proper exercise of Congress's spending power, and the Court held that, incident to that spending power, Congress could attach conditions to the receipt of federal funds. And so that case, you know, was really very different from what we're talking about here. And it’s a lot more complicated here—because the spending is supposed to promote the general welfare, as required by the Constitution, and the conditions imposed on the states have to be unambiguous. That was the upshot of the Dolecase.
Scott Caravello: Yeah, so then looking at the mechanism in the executive order itself, how does it work here when the White House is directing the Commerce Department to determine eligibility for the program?
Katherine Forrest: Well, Congress has already given the Commerce Department the authority to shape eligibility for—here, it's broadband funds—it’s not highway funds; it's broadband funds. So the executive order is tied to the relevant statutory provisions.
Scott Caravello: Got it, got it. Okay, so boiling all this down, what's our takeaway here?
Katherine Forrest: Well, the takeaway is that the White House is pulling on a number of levers to try and shape state-level AI regulation. And we haven't gotten into too much, but there's still a lot of opposition on both sides of the aisle to the concept of preemption and interference with state-level regulation—state laws.
Scott Caravello: And the states are still active.
Katherine Forrest: They really are very active right now. This is December 11th, and the executive order did not cause people just to go quiet. You know, last week, Governor Hochul of New York said that a ban on state AI regulation would be devastating to states. And then, on the same day that the President issued the executive order, she signed a bill that would force advertisers to disclose the use of AI to replace human actors. And there's also news reporting that she's proposing to change, you know, the New York RAISE Act, which is currently on her desk for signature, which would impose transparency requirements on frontier AI, similar to California's SB 53. So we'll see what happens. And the states are not quiet.
Scott Caravello: Yeah, and then also last week, as we wrap up, the 42 state and territorial attorneys general sent a letter to a bunch of AI developers demanding that they implement more safeguards to protect their residents and, specifically, children and vulnerable populations. So it just goes to show that this whole discussion about federal versus state AI regulation is far from over.
Katherine Forrest: You know, we're going to be reporting on this over the coming sort of months in the next year and seeing how these different provisions get implemented. But that's all we've got time for today. And we're going to be back next week, and we promise we'll talk about the second part of our interactive deepfakes—how to try to spot them when you see them. But we're signing off for today. I'm Katherine Forrest.
Scott Caravello: And I'm Scott Caravello. Don't forget to like and subscribe.
