Welcome to part two of our AI Policy Legal Primer. In this series, we ask a panel of leading appellate lawyers to assess the state of the law in determining the respective, lawful roles of the federal and state governments in regulating AI. If you missed part one, Preemption, Explained, check it out for a quick and helpful overview.
The dormant Commerce Clause has been anything but dormant in the last couple of weeks. With Congress and the administration actively debating the proper roles of the federal and state governments in regulating AI, the dormant Commerce Clause has emerged as an important topic of date.
The dormant Commerce Clause is a doctrine that prohibits states from passing some laws that unduly burden interstate commerce or that reach outside their borders, even in the absence of federal legislation. Although it leaves significant room for states to regulate, it does impose some limits on state governance, helping to ensure national markets remain unified, rather than fragmented by burdensome, extraterritorial, or discriminatory state rules.
We first wrote about the dormant Commerce Clause amidst numerous state AI proposals focused on regulating how AI models are built. In this piece, we reiterated what the Constitution outlines as the roles of the federal and state governments in regulating AI: Congress should govern the national AI market, and states should police harmful uses of AI within their borders. The dormant Commerce Clause isn’t an obstacle to all state AI regulation, but a guidepost for effective AI governance that respects the important roles of both Congress and the states.
In part two of our AI Policy Legal Primer, leading appellate lawyers Allon Kedem, Paul Mezzina, and William Jay are back to explain the dormant Commerce Clause and how it intersects with AI policy today. They discuss how courts use principles like extraterritoriality and tests like Pike balancing to weigh challenges, and examine what those frameworks could mean for recently enacted or pending state AI laws, including California’s SB 53, Colorado’s SB 205, and New York’s RAISE Act.
If you missed part one, check out Preemption, Explained. And stay tuned for a conversation on AI and the First Amendment.
This transcript has been edited lightly for readability.
Matt Perault (00:05)
So we’ve talked about how Congress can act to preempt. We’ve talked about limits of Congress’s ability to preempt. Paul, can we bring you in here? How does the Dormant Commerce Clause figure into this?
Paul Mezzina (00:16)
Sure, the dormant Commerce Clause is a limit on the states. And it’s called dormant because it’s something that the courts have taken by implication from the Commerce Clause.
The Commerce Clause is in Article 1 of the Constitution. It’s where Congress’s powers are listed, and it’s an affirmative grant of power to Congress. says Congress shall have the power to regulate commerce with foreign nations and among the states and with the Indian tribes. So it’s written in terms of what Congress can do, not that states can’t do. But despite that for a long time, going really back to the early 1800s, the Supreme Court has read the Commerce Clause as implicitly limiting state power. So the idea is when the framers authorized Congress to regulate interstate commerce, what they were trying to do was create a national marketplace where trade could flow freely across state lines and prevent states from interfering with that national market.
They intended for Congress’s power over national commerce to be exclusive. You can see this sort of going back to, there’s a case in the early 1800s called Gibbons versus Ogden, where New York was trying to create a monopoly on steamboat traffic. And that law was actually struck down as preempted. But in that case, Chief Justice Marshall said that the law might also violate the Commerce Clause because it interfered with Congress’s exclusive power to regulate interstate commerce.
Matt Perault (01:42)
Can you talk us through recent developments of the dormant Commerce Clause? I think there is, just in terms of how people are generally perceiving it, and if you’re paying a little bit of attention to recent Supreme Court cases, but not necessarily following them line by line, there’s sort of the perception that the dormant Commerce Clause is becoming increasingly dormant. Is that accurate?
Paul Mezzina (02:03)
Well, it’s certainly accurate that it has been a source of controversy and it’s been in flux a little bit. And I’m sure, we can talk about the pork producers case, but a big case that the Supreme Court decided a couple of years ago was a case where California had banned the sale within California of pork that came from other states that was from pigs that were confined and raised in ways that California defined as inhumane. And that law was challenged by pork producers from other states who said that California was interfering with interstate commerce, the national market for pork. And the Supreme Court ultimately, ultimately upheld the California law, but they did so in a very splintered decision. And that’s created, I think, a lot of confusion about exactly what the state of the law is on the Commerce Clause.
There are clearly different groups of justices, and they overlap in sometimes confusing ways, who have different views about what the role of the courts should be in reviewing state law under this dormant commerce clause doctrine, and also different views on various aspects of the doctrine and how it should apply.
Matt Perault (03:15)
What’s your best guess if there was a state law passed and someone was challenging it on Dormant Commerce Clause grounds, what’s your best guess at how a judge would review it? Like what are the various different tests they might use to evaluate the law’s constitution?
Paul Mezzina (03:30)
Yeah, sure. So I think the doctrine has basically three branches, speaking broadly.
The first branch is a rule against discrimination. And this is something that basically all the justices agree on. States cannot pass laws that discriminate against interstate commerce. So, you know, some examples of that from sort of historic cases. You know, there was a case where New Jersey banned out of state garbage from its landfills; said our landfills are only going to service New Jersey garbage. There’s a case where Hawaii gave tax breaks to liquor only if it was made from produce that was grown in Hawaii. Those kinds of discriminatory laws are invalid.
The second branch is one that is concerned with states regulating extraterritoriality. The idea is that a state can only regulate things that happen within its borders or that have a sufficient connection to the state. The state can’t reach out and regulate what’s going on in other states. And that’s a branch of the doctrine that really is, has been thrown a little bit into confusion after the pork producers decision. I think it still has some force where states are really reaching out and regulating outside of their borders, but exactly how much force that doctrine has after pork producers is a source of debate.
And then the third branch is what’s called Pike balancing. This comes from a case called Pike vs Bruce Church. And it’s basically a test where the court asks, is this state law imposing an undue burden on interstate commerce? And the court is supposed to weigh the burden that the law places on interstate commerce against the localized benefits of the law. And if the burdens are clearly excessive in comparison to the benefits, then the court is supposed to strike down the law.
Matt Perault (05:15)
I want to get into the controversy around extraterritoriality in a second. I would love all you to weigh in on that. But before we go there, can we just talk a little bit about Pike balancing? [Judges] do balancing tests all the time. I think when you hear about a cost benefit analysis, at least I sort of think in economist terms, I’m not an economist, but I think about that as an economics exercise. How would a judge go about trying to do that in a disciplined way?
Paul Mezzina (05:42)
Yeah, it’s very difficult. And this is why, you know, as we saw in the Pork case, there is a distinct minority of justices on the Supreme Court who think that courts shouldn’t be doing this at all. They say, you know, this is really a balancing exercise that requires judges to sort of compare incommensurable values, to weigh things that can’t really be weighed against each other in any legal way that requires courts to make value judgments.
For example, if the burden on interstate commerce is that it’s harder, it’s more expensive for businesses to engage in trade and commerce across state lines, but the localized benefits are potentially greater safety for state residents or maybe even serving the moral values of state residents, as in the case of that California law where California had a moral objection to the way certain pigs were being confined.
You know, it’s a really hard question how we’re judges supposed to determine whether, for example, California’s moral interest is outweighed by the economic burdens on commerce. And I, you know, we can talk about different ways that courts have approached that, but I think one way to think about it is that courts are really looking for red flags. They’re looking for cases where either the burdens on commerce are especially severe and disruptive [or where] the benefits to the local market are really speculative.
So cases where states are imposing really severe burdens for reasons that seem far-fetched or maybe even potentially pretextual. And just to take sort of an example, you know, there are some cases, historically, where states have imposed really problematic restrictions on, traffic through the state.
There’s a case where Iowa prohibited trucks from traveling on Iowa roads with 65 foot trailers and 65 foot trailers were basically the standard in all of the other states in the Midwest. But if you wanted to drive your trailer through Iowa, you had to either, you had to either not do it, navigate around and go through other states, or you had to transfer everything to a smaller truck while you went through Iowa. And Iowa said this has benefits in terms of local safety. And the court said, first of all, this is really burdensome. This is going to be really disruptive to commercial traffic throughout the Midwest. And also, we just don’t believe you about the safety benefits. We’ve looked at this. We think this is really far-fetched. There’s really no clear benefit. 65-foot trailers aren’t any more dangerous than 55-foot trailers, which Iowa permitted. And so those kinds of laws where the balance really seems out of whack are the ones that courts tend to be comfortable striking down.
Matt Perault (08:26)
So I want to, I think your truck example is making me want to get to an application of AI. How do we think about these laws in an AI context? But before we get there, Paul, you did allude to this controversy around how much weight to put on the extraterritorial prong versus potentially on others. wrote a piece on dormant commerce clause, where we focused primarily on Pike balancing and we looked a little bit at extraterritorial effects, but within the Pike balancing context.
Can you talk a little bit about how these two different prongs might be weighted? Like another law professor, Kevin Frazier, at the University of Texas has put a lot of weight on the extraterritoriality prong, not just from the dormant Commerce Clause, but from other constitutional principles. So how much of that survives and how would that be relevant to thinking about analysis in the AI?
Allon Kedem (09:10)
So it’s a little bit unclear how much survives and therefore exactly how it would apply. When we’re talking about extraterritoriality, we’re talking about one state’s attempt to essentially regulate conduct or commerce or something that takes place in another state. And it’s often not something that can sort of cleanly be divided into what happens in one state versus another. So you could take the example of a transaction that involves a you know, let’s say a sale between, you know, someone in Iowa selling a good made in Minnesota to someone in Maine. And the question is, what state does that transaction take place in? It’s a little bit hard to know. Or in some instances, you could have the use of a product in one state that has spillover effects into another. And so it’s not actually clear in that instance that you would necessarily be regulating the out of state conduct as opposed to the causing of effects within the state.
And for that reason, I think it doomed the argument in California’s case because of the pork producer challengers, because really what the California law applied to was the sale within California of meat that had certain characteristics.
Now, it’s true that in order to comply with the law, if you were an out of state producer, you might have to take certain steps outside the state. But nevertheless, the thing that actually puts you on the wrong side of the law happened wholly within the state of California.
You can contrast that to an instance where a state passed a law that says essentially you can’t take some step abroad. And let’s say they tried to use as an in-state hook the fact that the company just is registered to do business within the state or has employees within the state. But nevertheless, the thing that is being regulated is the out-of-state conduct. And in that instance, I think you would still have some vulnerability under the anti-extraterritoriality principle. That said, the majority in pork producers suggested, although didn’t squarely hold, that the only types of laws that are per se invalid under the principle are laws that essentially tie the price of goods in one state to the price of goods in another state.
So there were a number of laws primarily in the area of liquor sales where one state would essentially force a liquor seller to affirm that the prices made available within the state are no higher than they are made available in other neighboring states. And the Supreme Court struck those down. And those rulings are still good law, but the pork producers majority suggested that it’s possible that other laws that don’t have to do expressly with tying might not get the same treatment.
But one caveat is at the same time it said, even if those laws aren’t invalid under the dormant Commerce Clause, they still nevertheless might run afoul of other constitutional principles, including the separation of powers and the sort of inherent limitations that each state’s sovereignty imposes on the sovereignty of a sister state. And so it would infringe the sovereignty of Iowa, for instance, if Florida were to enact a law saying that people are not allowed to engage in certain conduct within Iowa because that would be an instance of Florida projecting its own sovereign authority outside its own state
Paul Mezzina (12:30)
I’ll say the way I think about it, and I think there are different views on this, and it’s not clear yet where courts are going to come out. I think one clear takeaway from the pork producers case is that a state law is not going to be invalid just because it has practical effects outside of the state.
So in pork producers, what California was regulating was the sale within the state of pork had certain characteristics that came from pigs that were treated in certain ways. But the object of regulation was the in-state sale. And the farmers were saying, yeah, you’re regulating the in-state sale, but that’s going to have a whole lot of effects on our business in Iowa and North Carolina and other states. And the court said those kinds of out-of-state practical effects just aren’t cognizable under the Commerce Clause. We’re not concerned about that.
I think on the other hand, what is still open after pork producers is when a state is actually directly regulating what happens outside of the state. So for example, if you imagine sort of a tweak on the law in the pork producers case, if instead of California saying, it is illegal to sell pork in California, if it comes from pigs that were treated in a certain way, suppose California said it is illegal to raise pigs in an inhumane way if that pork later makes its way into California.
In that case, the object of the regulation is the way the pigs are being raised in, for example, Iowa. And it still has a connection to California. But I think that’s a fundamentally different mode of regulation. And that, I think, still raises some serious questions under the Commerce Clause.
Allon Kedem (14:17)
One other thing that the Supreme Court emphasized in the pork producers case is that if there is a way to essentially change your method of production or your way of doing business in order to comply with an enacting state law, then it severely undercuts and in some instances renders impossible a dormant Commerce Clause argument. And in that case, one of the things the court pointed out is that you could simply have a different method of production for California or just choose not to sell in the state whatsoever. And therefore, it was essentially up to the pork producers themselves to figure out how to comply with the law and that engaging in commerce in a particular way, in a particular preferred manner is not something that the dormant Commerce Clause.
Paul Mezzina (15:04)
Well, just to pick up on that, because I’m interested in my fellow panelists’ views on this, I think the point that Allon just made, as I understand it, is a view that was only taken by four justices and was disagreed with by five justices. But despite that, under what’s called the Marks principle, which is supposed to help you figure out when the justices are divided, what is the controlling rationale that lower courts are supposed to follow, there’s an argument that that that rationale that said there’s there’s no burden on interstate commerce if you can change your your method of doing business is controlling even though a majority of the court disagreed with it.
So it puts you in a tricky position if you’re a lower court judge. You know, am I supposed to follow this minority view because it was technically controlling or should I take into account the fact that five justices seem to have disagreed with?
William Jay (16:02)
Just one thing to observe is that some justices really don’t like the dormant Commerce Clause, but support for the idea that one state can’t regulate things in other states might not be limited to the dormant Commerce Clause. There are some alternative theories, there are some lower court cases, and even some older Supreme Court cases that spin out some alternative theories like under the Due Process Clause. We’re going to spend most of the day talking about the dormant Commerce Clause but there are some alternatives that a litigant would probably keep in their back pocket.
Matt Perault (16:31)
Can you say a little bit more about that, the due process clause? Can you explain the rationale there and then [what other] hooks to draw on beyond the Due Process clause?
William Jay (16:40)
The Due Process clause, I think, is the main one, the, you know, the whatever Justice Gorsuch meant when he wrote in the pork producers case about the horizontal separation of powers, like that’s not a specific provision of the Constitution. It might be a reference to a whole bunch of different aspects of the Constitution that keep states in their own lanes. But the one on which there is the most case law is the due process clause, which is the idea that you can’t be deprived of life, liberty and property without due process. Having a state punish you, taking away your liberty or property for something that the state has no valid connection with. If you’ve never been to Idaho, but Idaho passes a law that says if anyone who goes from Virginia to England and drinks a pint and eats stinky cheese, shall pay a fine to the state of Idaho. Idaho’s got no connection with that travel and it would violate due process to impose that kind of punishment. There have been a handful of examples often involving some international angle and not just interstate angle in the past. The Supreme Court hasn’t had to reach that in a long time.
Matt Perault (17:59)
So I think it’s good we spent as long as we have on the foundation, because it kind of gives us the analytical lens to now look at state AI laws. So I’d love to get your thoughts on some of the ones that we’ve seen.
I’ll list three, two of them have been passed, one of them is pending. And Willy maybe we can start with you to give some thoughts on, what are the kinds of elements of these laws that raise concerns and which are the things maybe that are more likely to survive a potential challenge?
So one is California’s SB 53, which is a set of disclosure requirements for model developers. There’s Colorado’s SB 205, which was passed a couple of years ago, is now set to go into effect, I believe, this summer. And the legislature has been actively considering whether to revise it in some way. There’s been pressure from the governor and the attorney general and others to reconsider it in some form. That law is primarily focused on algorithmic discrimination. Neither the California law nor the Colorado law include any provision limiting the scope of the law to development or deployment within the state. So I’m curious how you guys would assess that. And then the final one is the RAISE Act in New York, which hasn’t been signed into law, but passed the legislature. It’s on the governor’s desk for signature. And that would require developers to implement safety and security protocols, to describe in detail the testing procedure they use to assess a model’s potential for harm, and then to implement safeguards to prevent unreasonable risk of critical harm.
Willy, starting with you, what are the things in these laws that give rise to more of the kinds of concerns that might be taken into account in a dormant commerce clause challenge?
William Jay (19:36)
There are some examples in, I think, more than one of these laws that say, for example, until you have filed a set of disclosures with our state, you may not market a model that has been developed or deployed in certain ways. In other words, it purports to just be a disclosure requirement, but the teeth in the disclosure requirement are that you’re not supposed to be deploying the model anywhere, including outside the state with the state law until you’ve filed the paperwork in the state.
And while that may not seem like a very large burden, and if we were just balancing benefits and burdens, that might not be a huge obstacle. It’s still sort of the camel’s nose into the tent, that you need New York or California’s permission to go and do things in the interstate or even international market.
Allon Kedem (20:44)
And Matt, one of the things to pay attention to with these laws is whether they specifically articulate a geographic reach for the law. And if they don’t, how those states’ courts would interpret the laws, because a lot of times you’ll have a law that’s sort of stated in general terms. No one is allowed to do something to engage in some activity. But nevertheless, in order to avoid dormant Commerce Clause and other constitutional concerns, the law might be interpreted as applying only to the product as deployed or used or sold within the state, as opposed to its sale, deployment, or development outside the state. And so there’s a question, I think, for at least some of those state laws, whether they are genuinely intended to reach out-of-state conduct, even if it’s not connected to any sort of direct in-state commerce.
Matt Perault (21:37)
I think this raises a really good question because there’s the question, what’s written in the law and then the realities of AI development and deployment and how those two intersect. So if you take New York, which is, I think the language is roughly developed or deployed in whole or in part within New York. then, so that is clearly a provision that limits the scope of the legislation.
But I’m curious how you think that would work with the realities of AI development. So you might have an open source developer in California, for instance, [who’s going to develop] the technology in California, but then have no control sort of by definition over its deployment, including this potential deployment in New York. You can have the reverse as well. How would you think about the realities of AI development against the provisions, a jurisdictional provision in a bill like RAISE?
Allon Kedem (22:28)
And I think this goes back to the point we were just discussing in the pork producers case about whether you can segregate your production into different markets. So in the pork producers instance, at least theoretically, it was possible for pork producers to say here is meat that is destined for the California market that satisfies California standard. And then here is a stream of production for all the other states. And there’s a question in the instance of AI technology whether you can have that same sort of segregation of markets.
In some instances, it might be possible through geofencing or other technologies to say you just can’t access the website if you’re trying to do it from an IP location in a state that has a ban on it. In other instances, it might not. If you’re talking about open source technology that essentially can be used and deployed by anyone redeployed, used, incorporated into technology downstream it may be that the AI developer loses any ability to segregate into different geographical streams.
Matt Perault (23:31)
Paul, how do you see it?
Paul Mezzina (23:34)
Yeah, I agree with that. I think a really important question that courts ask when they’re thinking about the Commerce Clause and also when they’re thinking just about personal jurisdiction is, can you be essentially found in violation of New York law without having directed your conduct toward New York?
So if you look at how courts have thought about state regulation of the internet over time, in the early days of the internet, there were some cases where courts said, you know, Vermont cannot regulate what’s on a blog or a webpage because that’s just going out to the whole world. And we can’t, if we let Vermont regulate it, that Vermont will be controlling conduct in all of the other states. And then Allon mentioned geofencing. As geofencing technology became better, you started to see courts saying, for example, there was a case involving a California law that required CNN to put closed captions on videos on its website. And CNN challenged that out of the Commerce Clause and California said, it’s fine because you can detect, you can use technology to identify when someone from California is accessing your website and you can just provide closed captions to those people and you don’t have to provide closed captions in other states.
So I think that is going to be a really important part of how courts look at AI regulation, is they’re going to ask, is this state law only going to apply to AI developers or deployers who are actively engaged in conduct within the state or directing their conduct to the state? Or is it potentially going to reach out and affect open source developers and developers who are not actually directing their conduct toward the regulating state and are just being pulled in because their product through the stream of commerce somehow made it to the state.
This newsletter is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. Furthermore, this content is not investment advice, nor is it intended for use by any investors or prospective investors in any a16z funds. This newsletter may link to other websites or contain other information obtained from third-party sources - a16z has not independently verified nor makes any representations about the current or enduring accuracy of such information. If this content includes third-party advertisements, a16z has not reviewed such advertisements and does not endorse any advertising content or related companies contained therein. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z; visit https://a16z.com/investment-list/ for a full list of investments. Other important information can be found at a16z.com/disclosures. You’re receiving this newsletter since you opted in earlier; if you would like to opt out of future newsletters you may unsubscribe immediately.
