Chairman Ferguson, thank you so much for sharing your insights with my University of Texas School of Law students.
I'm really, really happy to be here and congrats on getting over to UT. It's a great school. Austin's a great town.
You know, we like to say that the future arrives early in Austin, so I'm glad to be there and I'll let you know what I see from the future, but right now I want to focus on what you're experiencing right now because the FTC is doing an incredible amount of work in the emerging tech space. But for folks who have been under a consumer protection and/or competition rock, let's just start with the FTC's mandate on those kind of two prongs. What is the obligation and enforcement power of the FTC with respect to consumer protection and then we'll move into competition?
So we enforce a lot of laws, but the most important law we enforce is Section 5 of the FTC Act. Congress adopted the FTC Act in 1914 because they were dissatisfied both with the Sherman Act, our original antitrust law, and with the Department of Justice's enforcement of the Sherman Act up to that date. So 1914, they adopted the Clayton Act, which if you've taken antitrust you'll know about that one too. And at the same time, or shortly before actually, they adopted the FTC Act which added a new agency to enforce the antitrust laws and a new prohibition in antitrust, a prohibition against unfair methods of competition. We had about 20-ish years of enforcement under that. And then in the late 1920s the Supreme Court in response to the FTC going after basically deceptive advertising said Congress made this about competition. So unless you can show that the deceptive advertising is injuring competitors rather than consumers, you can't use unfair methods of competition prohibitions to pursue them. And so Congress in the mid-1930s adopted the Wheeler-Lea Act or the Wheeler-Lea amendment which added arguably the most important part of the FTC Act, which is an additional prohibition against unfair or deceptive acts and practices.
And deceptive is sort of easy probably for your law students if they took first year torts know about fraud. And so deception sort of sounds in the old torts of deceit, the torts of fraud. It's a pretty straightforward prohibition. If you lie to someone in order to get them to buy a product or service you are committing an act of deception. One of the most important things, however, was that Congress used deceptive acts or practices and not fraud or deceit because one of the things that Congress made clear, and this is laced throughout the legislative history for those who care about that type of thing, that they were attempting to relieve the American consumer of the burden of caveat emptor. So, our normal rule is as long as no one actively lied to you about something that you're buying, you're sort of on your own. You have to test it out and assure yourself that the thing is what you believe it is. And as long as they didn't tell you a lie in order to induce you to enter into some sort of transaction, fraud and deceit generally sort of stand aside. The FTC Act also prohibits with that phrase deceptive material omissions. And the way to think about that, which our caveat emptor rule generally would say no, like if they omitted something, it's sort of on you to figure it out and then to write a contract that insulates yourself to the best you can about potential omissions. And so the omissions prohibition is against material omissions. And the way to think about that is that the FTC Act bans both active deception where someone tells a categorical affirmative lie, but also failure to disclose something that a reasonable consumer would want to know before they entered into a transaction. And the sum of that prohibition on deception, affirmative and omissive, is that sellers are required to act consistently with reasonable consumer expectations. So when we bring enforcement actions, the sort of abstract question we're asking ourselves is what would a reasonable consumer be understanding is happening in this transaction? And if they say something designed to elude the reasonable consumer's expectations or if they omit something that that reasonable consumer would obviously have wanted to know about the transaction, it's a Section 5 problem. Then the second prong we have is the most controversial prong of that new Wheeler-Lea.
And before we get too controversial, let's hang our hat for a second on this reasonableness and bring it to AI for a second. Because when we think about material omissions, when we think about some degree of perception, if we think about that in the context of a car, I think we can all imagine what we would like to know. We would like to know how many tires it has, how many doors it has.
Is it a lemon? Right? But we're still figuring out what a lemon means in the AI context, right? Because this is something that is new. It's nondeterministic. It has all these strange features. I don't know if we can find a quote-unquote reasonable view of AI with respect to the average American consumer. So, in these emerging tech spaces, how do you begin to think through what reasonableness means? I know this is a question that myriad law professors have tried to answer in hundreds of pages of law review articles. So, if you could make a lot, it'll make a lot of people tenure, I'm sure.
Yeah. So most of our AI work has been at the application layer. And so if you sort of think about AI as having three layers, it's the infrastructure layer, data centers, chips, power, water, etc., the sort of foundational model layer, think like Grok, ChatGPT, Claude, etc., Google Gemini, and then there's the application layer which are a huge range of applications that use the foundational models or specialized models to do specific things for consumers who want to do them. Like a great example that all of your students will be dealing with when they leave law school if they go work in the private sector is all sorts of AI tools to do document discovery to process terabytes and terabytes of information that would have taken warehouses of lawyers often abroad weeks or months to do that these AI applications will be able to do in hours or days.
We're already using Harvey. We've got... Yeah. No, we're... I told you the future arrives early in Austin. We're leaning in.
Yeah. I'm an old millennial. I'm still trying to catch up. And so most of our work has been at the application layer as opposed to the foundational model layer. And then at the infrastructure layer, we've done competition work in that space. And we can talk about that later. But at the application layer, at the end of the day, these apps are just like any other app that existed before and just like a huge range of products and services that existed before AI, which is someone created something. They're going to try to convince a consumer that it's capable of doing certain things and it either is or it isn't. Or as is the case of AI and actually is the case of a lot of health claims that we've been litigating since the 1930s, maybe it can do the thing but when you made the claim you couldn't substantiate that you knew that it could do that thing. You didn't have evidence, like you had a hunch maybe, but you didn't know for sure. And that's an additional violation of the FTC Act is if you make a claim about something but can't substantiate the claim that you're making even if we can't prove that it's categorically untrue if you can't demonstrate you had evidence that it's true that's an FTC Act violation. And so we've brought cases both in the last administration and in this administration where some sort of AI tool was being sold as capable of the following things. It can do this and it can do that. And our investigation revealed that it couldn't do that and they knew it couldn't do that. And that is no different than any other sort of fraud that we would bring in any other industry. Someone marketed a product as being able to do certain things or to achieve certain results. It couldn't. And so we bring a deception claim.
And you pointed this out very clearly when you were commissioner under the prior administration thinking about Rytr. Rytr was a tool that was capable of producing myriad consumer reviews of products. There was an argument that it was facilitating, it was a means and instrumentality to illegal actions and therefore maybe was something that should be targeted. But as you wrote in a dissent, hey there is no AI exception to the law. We don't need to treat this as something that's abnormal. We can just adhere to what we've been doing the whole time.
Yeah. And so that's exactly what I said. And it's also important to think about. I mean, look, like hammers have millions of legal uses and also some illegal uses. I prosecuted a case when I was solicitor general against a guy who was sentenced to life for using a claw hammer to paralyze someone in the commission of a crime. So hammers can be used for horrifying...
I thought this was merely going to be just a metaphor, but you really...
No, no, I actually did. The hammer one is the one that sticks in my head because I defended the conviction in this case. But like paper, pencils, pens, all of which can be used and are used routinely to do perfectly lawful things can also be put to fraudulent uses. And no one would think to prosecute a paper mill because its paper ended up being used to commit fraud or far worse crimes. And my view in Rytr was yeah okay that this tool is undoubtedly capable of facilitating fraud. We actually didn't have any examples of it being used to facilitate fraud which I thought was one of the biggest problems with this theory that we had no real world examples at all. Commission staff had generated some fraudulent ones to reveal that it was capable of that. But if this tool was capable of myriad acts of fraud we would expect it to find some in the real world and we did not. And so that was my first objection was like this feels like a solution in search of a problem. But the other was we have in IP law, in fact, the court decided one of these cases just a month ago that says, yeah, look, if there's a tool capable of copyright infringements, that alone does not make the tool's maker liable for copyright infringements, unless you have evidence that the tool's maker knew and intended that the product be used for those infringements, as opposed to aiming it for lawful purposes, even if people used it for a lot of unlawful purposes. And that was my view here is like we have this pretty long legal tradition to handle these multi-use products where some of the uses are illegal and some of them are legal. And because of the innovative entrepreneurial pioneering spirit in America, we don't take what feels to me like a very European assumption which is if I can demonstrate it's capable of illegal uses, we ought to regulate or ban it. And we take the opposite view which is as long as they weren't making it for that purpose or they weren't aware that it was being used exclusively for that purpose and continued to make it then we let it go and that was my view in Rytr and it's why we undid the Rytr decree when I became chairman. The president in one of his AI orders told the agencies like go back and review things the Biden administration which was very hostile to AI innovation see what they did and if they did anything that was illegal, you know, was not itself lawful or inconsistent with our favorable attitude toward measured AI regulation, you should undo it to make sure that we're not getting in the way of innovation. And Rytr was one of the first ones we did. That was easy. I had dissented, Commissioner Holyoak, now US Attorney Holyoak had dissented. It was a pretty long dissent where I laid out like the entire history of the means and instrumentalities theory going back to the very early 1920s on the commission. But yeah and I think that is an example of this administration's sort of change in tone and posture which is don't assume the worst of these new and emerging technologies. Confront problems where they arise but do not assume that the new thing is likely to present a problem. And I think that the watch word here is Europe because that is and has been the European approach. Europe has not come close to innovating in the space at all. No one is founding companies there on this front. Anyone who wants to found moves from Europe to the United States. And it is not because they don't have rule of law. They generally do. It's not because they don't have safe streets and established infrastructure. They do. They have a regulatory climate that makes it impossible to innovate at the pace you need to innovate to keep up with innovation elsewhere. And a regulatory climate that is presumptively hostile to novelty, which is generally the opposite of the approach we've taken in America.
Yeah. And you emphasized in some remarks earlier this year that it's not the FTC's role when there's trust, when there's reliability to step in. In fact, that's when you all need to get out of the way. And when you think about what many AI labs are actively pursuing right now, doing a lot of safety work, doing a lot of transparency work, trying to disclose that to allow for that sort of competition. Well, now, going back to the car analogy, if you see one car manufacturer say, we did this crash test safety rating. It's this safe, so on and so forth, and then another car manufacturer just says, we think it's safe. Consumers are going to naturally gravitate to the one that has more transparency that's being open. And so that didn't require a new regulation. That didn't require specific law. That was just working off of what reasonable people want and are looking for.
Yeah. And look, the president rolled out his sort of national framework for AI policy and regulation a couple weeks ago. And it isn't just sort of openly libertarian, in fact it's not openly libertarian at all. They have the sort of four C's framework, have to protect children, communities, creators, whatever the fourth C is, but like a lot of that is affirmative regulation aimed at promoting innovation but not allowing an industry to run totally amok, preserving competition. And also I think the president's instinct here he has said it a lot and I share it is if we're going to do this well we need a national regulatory framework because of the risk of the sort of balkanized state regulatory patchwork which we've seen for example on the privacy side. I think the absence of federal privacy legislation has been a problem both for the development of the internet and also for the development of AI. And it's been extremely frustrating for consumers. And I think that if we don't institute a carefully thought out national framework for AI regulation, we're going to end up with one of two things. A 50-state patchwork where regulatory compliance ends up kind of looking like Europe, which we don't want to have, or California as sort of the big player as it is on lots of markets ends up setting all the baseline rules and then you have one part of the country getting to dictate the rules for the rest of it rather than the nation's representatives gathering in the capital on an issue undoubtedly of interstate commerce, sort of the core of what the framers wanted Congress to address on matters of economic regulation and set an affirmative framework that both lays out pro-innovation ground rules that protect communities, kids, creators and make sure that rate electricity ratepayers are not sort of footing the bill for the expansion of AI infrastructure. But also ensures that that is the lone framework and that the states don't get to either layer on top of that a huge additional patchwork of regulations or that there's no national framework and all we have is a patchwork of regulations.
And I think it's particularly telling to note that this California effect has been in play in other industries as you noted but those industries have generally been established. We've had decades before California intervened and said hey this is the way of the road, literally cars.
Your cars. Yeah, that's another good example.
