Sen. Adam Schiff on AI regulation, free speech, and impeaching Trump one more time

extract

Today, I’m talking with Sen. Adam Schiff, a Democrat from California. Schiff sits on a number of committees with oversight into tech and AI: intellectual property, antitrust, privacy and technology — it’s all there. I really wanted to ask him about how we might regulate anything related to the tech industry at this moment in time. Sen. Adam Schiff on AI regulation, free speech, and impeaching Trump one more time Congress missed its chance on social media. Can it act on AI? Verge subscribers, don’t forget you get exclusive access to ad-free Decoder wherever you get your podcasts. Head here. Not a subscriber? You can sign up here. But as you’ll hear, he started our conversation by talking about the self-dealing and corruption present all through our politics. That of course fell against the backdrop of President Donald Trump gathering AI CEOs to the White House to sign a nonbinding pact in which they agree to, you know, do a good job keeping AI safe. I asked Schiff if he thinks that’s enough, how he thinks a more formal regulatory framework would look, and what might happen next if Democrats retake Congress in the midterms — all the way to whether he’d support impeaching the president (again) over all that corruption. We also spent some time comparing and contrasting how you might regulate AI with where we’ve landed regulating social media. One nerdy note: You’re going to hear me ask the senator about something called Chevron deference and a Supreme Court case called Loper Bright. Basically, the Supreme Court overturned a Reagan-era precedent saying courts would defer to experts in agencies when the law was vague. Now, the courts get to decide, which could cause any new regulation to get immediately tied up in litigation. That seems awfully relevant to the idea of creating a new agency to regulate AI, which Schiff supports doing. Okay: California Sen. Adam Schiff. Here we go. This interview has been lightly edited for length and clarity. Sen. Adam Schiff, you’re a United States senator from California. Welcome to Decoder. Thank you. Great to be with you. I have been asking every politician who comes on the show this very basic question here in 2026. Senator, what the hell is going on in our country? It’s a very good question. I think we have seen things take place in the last couple of years in particular that we’ve never seen before. There’s sadly rampant corruption within the administration. Every day, every week, it’s a new thing. At the moment, I think what is most jarring is seeing campaign ads aired of the president that are paid for by taxpayer dollars. But every week it’s something else. It’s excluding press from the White House or Air Force One, or it’s some new moneymaking scheme or some new vanity project. If you look at, for example, the triumphal arch, they’re now claiming it’s a national security issue because they want to store ammunition in it. And why do they want to do that? Because the best legal argument they have is that this is not just cosmetic, that they have some national security purpose. So what are we seeing in the country? We’re seeing chaos and no focus on the cost of living and improving the quality of people’s lives at the administration level, or by my Republican colleagues in Congress, but I’m hoping that changes. The national security piece is very interesting to me in the context of a bunch of questions I have for you, because the Trump administration loves to justify every single thing it does by saying it’s national security, and it can’t be questioned, and this should bring more authority to the executive branch in one way or another. And they’ve been doing that since the beginning. But just this week, we had a bunch of tech executives at the White House signing a non-binding promise to evaluate their own AI systems. And there was no national security concern that made any of that binding, even though they are among the most prominent people in the country saying they might kill everyone. Can you square that circle at all? It’s very hard to square. It was interesting to see all those tech leaders there signing onto this voluntary pledge on the one hand and on the other hand, away from the White House grounds, away from the president, being very public and saying, “We need regulation. We don’t want this to be just ‘Trust us, we’ll do the right thing.’” So yeah, it was very distant, very jarring. None of these companies want to incur the wrath of this very vengeful president. Apparently, they will say what he wants them to say in his presence. But the reality is, there is a real national security threat here. As these models become more and more advanced, as they advance themselves through this recursive AI, we have less and less of a sense of what they’re doing, how they’re deciding what they’re deciding, and what they’re trying to access lawfully or unlawfully. That is a national security concern of the highest order. There are other security concerns too. The unbridled implementation of AI at the Pentagon can result in calamities, and we need to investigate what role AI may have played in target selection in the bombing of that Iranian girls’ school, for example. There has been an effort to go after Anthropic, because Anthropic wanted to insist on not using AI for domestic mass surveillance or for fully autonomous weapons systems. That’s a national security issue. And their response in the administration was to try to kill Anthropic, and they’re still at it. So lots of national security concerns, none of them aired. When the president was asked about this by a reporter, he said, “Well, the world doesn’t have to worry about this because these companies love America.” I’m not really sure that’s much of an answer. And when he was asked, “Well, isn’t there anything else that’s necessary?” He pointed to his brain and wanted people to take comfort that he has, I guess, a very big brain, he tells us. He’s a stable genius, and therefore we don’t need to worry. He’s passed all the tests. So I’ve been told. You are obviously on a number of committees, but most importantly, I think for this conversation, you’re on a number of subcommittees as part of the Senate Judiciary Committee that are in the strike zone for AI. So you’re the ranking member on the subcommittee on Intellectual Property, you’re on the Privacy, Technology and the Law subcommittee, and then, of course, the Antitrust Competition Policy and Consumer Rights Committee. I just look at that list, and that is the dead center of the AI regulatory debate. What’s the conversation in Congress like as you debate these issues? In particular, I look at the IP issues, and they seem existential to these companies. If we start charging them for training data, they might not exist anymore. I think that’s right. When you look at, for example, all of the IP that’s been vacuumed up by these models to train these models, that’s a lot of the value of some of these entertainment companies. And so they live or die on the strength of their IP. But what makes this challenging is that not all of the players in the industry are the same. You’ve got big streamers involved; you’ve got big technology companies involved. You’ve got companies whose AI or their technology portfolio is much more important to them than the IP value of their film library. I don’t think the film library at Amazon is Amazon’s most important business. If it was, they probably wouldn’t have made the movie Melania. But my favorite review of that, by the way, was the Variety reviewer who said, “If you watch that movie on a plane, you would still walk out.” But the IP issues are really important. Senator Curtis and I are working together on them to begin by saying, we want to know what copyrighted material, what protected material, was used to train these models. And then we can have a discussion about what’s fair use and what isn’t, but we need these companies to retain that information. But lots of other IP issues, some beneficial. We had a hearing on the patent office, and the patent office’s use of AI to more quickly get through the backlog of patent applications. So there are going to be good applications too in the IP area, but also a lot of worry about what it means for people’s creative talents. I wanted to start with IP, not just because I’m a failed former copyright lawyer myself, and emphasis on the failed. I was no good at it. But it just seems like what this industry needs right now is a pretty holistic approach to regulation. So are the inputs of the models being collected in a legal way that benefits the ecosystem that created them? Are the studios going to get paid for a bunch of training data? Are website operators going to get paid for training Google’s models that they scrape with search? And then you’re on to, do they need an antitrust exemption so they can coordinate on safety and they don’t kill us all? And there’s a whole list of things in between those two ideas that also need some regulatory frameworks, that might just need some basic transparency laws. What do you think about that? Is our government prepared to start at the beginning, and end at the end, and do all the steps in the middle? Well, how I think about it is you’ve got all the input issues, and it really seems like the thought at these companies was, “We’re in a race. We’re just going to vacuum it all up. And yes, this probably violates a lot of the IP laws, but if we don’t do it, the other companies are going to do it. We’ll worry about the legality later.” So that’s what they’ve done. And of course they’re being sued by different parties, and some are settling these cases, and some are fighting these cases. But they began the process by just deciding they were going to ignore IP law, and they were just going to take [IP] and worry about the problems later. And then, in terms of the outputs, it has taken us to where we are today for these companies, I think, to consider the safety of the outputs. Do we need to start slowing down the pace of our development of these models because we don’t trust the outputs? We don’t know where they’ll lead. We can’t even control these models ourselves, let alone provide these models to the public, to good actors, to bad actors, and expect that we’re not going to have some terrible result. In terms of how I think about this holistically as a legislator, this is crying out for legislation. I am, I’ll be honest, very circumspect about Congress and legislating the tech sector because we did such a — I don’t know what language I’m allowed to use here — but such a shitty job when it comes to any kind of regulation of the harms, the societal harms coming out of social media. We’re going to have to do a much better job and much more quickly when it comes to AI, because the threats are even more significant and we’re already at the point of strong potential harms. It can’t wait. We certainly can’t take the kind of years we’ve taken to try to even wrap ourselves around social media. Can I ask about that comparison? I hear it a lot. I hear it from politicians. I hear it from the industry itself. I hear it from parents, of course. It seems to me that the problem with social media regulation was never that we’re too slow, or that no one can come up with an idea; it was always the very existence of the First Amendment. The thing you needed to do with social media was regulate the flow of disinformation, or keep teens from bullying one another, or from posting pro-anorexia content. All of that came down to being speech regulation. The companies knew this; the tech companies absolutely knew this. And I remember they would put up billboards in DC. Meta would put up billboards saying, “We welcome regulation.” Then I would listen to the ideas for regulation, and they would crash headfirst into the First Amendment, and then nothing would happen. And it seemed like that paralyzed us from doing things like privacy regulation, because it was tied up next to the very idea of social media. It doesn’t feel that way with AI. Do you and your peers in Congress draw the distinction that it isn’t the First Amendment preventing you from doing a lot of the things that you might do for social media, in the AI case? Honestly, I don’t think it was the First Amendment that was the issue in regulating social media. I think the First Amendment was certainly used by companies that didn’t want regulation to try to argue everything was a First Amendment issue, when so much wasn’t a First Amendment issue. The threshold question in terms of regulation of the social media companies is: is the broad immunity they have under Section 230 still working? The industry asked for that one-size-doesn’t-fit-all immunity because it said, “We’re a nascent industry, don’t regulate us, don’t inhibit us. If you try to regulate us, then we will feel, or if we’re not immune, that it’s necessary to take down content because we might get sued over it. We want to be good actors. We want to take down offensive conduct, but we don’t want to get sued for that either. So give us carte blanche.” And they got it. They got carte blanche, and they have not been good citizens, corporate citizens, and they’ve allowed a lot of harm to take place, and they’ve trained their algorithms for addiction. So much of the content has been harmful to young people, and so much of the content has violated their own policies, but they have been immune from any consequences because of the breadth of what Congress gave them. Now they found it, I think, strategically, a good strategy is to frame these things as the First Amendment, and to attack the other side and say, “Oh, there’s a bias that is against conservatives,” or whatever. It was all in the service of keeping that unregulated. So I really don’t think the First Amendment has been the issue, but I think it has been utilized as a bludgeon to try to stop any kind of meaningful regulation. I similarly worry that all of the money around AI may result in the same paralysis if members of Congress aren’t willing to do the right thing, and aren’t willing to put in place safeguards regardless of how much different industry players might spend. We’ve had Sen. [Ron] Wyden on the show many times to talk about Section 230. I won’t go too far down that rabbit hole. The dynamic there was that Section 230 provided the immunity that you’re describing, but it was immunity for speech that would otherwise be illegal. And so it was speech that was defamatory. There’s a whole category of speech that isn’t illegal. Are you saying that taking the 230 immunity away from the social platforms would somehow make them moderate lawful but awful speech? And I bring this up because I’m very curious about how you would apply it in the AI case as well. Yeah, no, I think that is the case. That part of their argument was, “ We want to take down really nasty, offensive, racist content, but we will be afraid to do so if you don’t give us immunity. So give us immunity, and we’ll take all that bad stuff down.” They got immunity. And I don’t know if you’ve been on X lately; it’s a cesspool. So giving them broad immunity did not allow them to enforce their community policies. And the First Amendment prohibits the government from dictating speech or prohibiting you from saying things. It doesn’t prohibit a private party from saying, “We don’t want racist content on our platform.” So there’s no problem with a private platform saying, “That violates our policy. We’re not going to post it.” And they’re not utilizing that immunity in a good way. So why should they continue to have it? Now, we may want to create a safe harbor for companies that are meeting best practices and standards. So we may want to narrow 230 instead of getting rid of it. What is the analogy in terms of AI? I’m not sure what the analogy is. Let me draw you the connection that I’ve been thinking about a lot. In the case of social media and Section 230, the immunity was that the platforms would not be responsible for the speech of the users. So I would post something horrible about you, and you couldn’t sue Mark Zuckerberg. This is the thing that has broadly enabled the social media platforms at scale. I opened Instagram today, and it is all but making the content for me. And in fact, a lot of content on Instagram is being generated by Meta. It is obviously their speech. And somewhere between having your Muse go and make Instagram for you, and Meta synthesizing all of the product recommendations and generating content all by itself, is a liability for Meta. It is outside of the Section 230 shield. That is their speech. And you can draw a lot of distinctions there. If I fully automate my Instagram account with my Meta agent, is that my speech, Meta speech, or the AI model’s speech? Who knows? But it might not be under 230. So you would be outside of the immunity shield. And I’m just curious if you think that will meaningfully alter how they moderate. I agree with your analysis, but I think it’s a difficult enough legal analysis that they feel they’ve got the money to stop, or slow down, effective litigation. Meta, or any of these other companies, decides how they want to amplify certain content. So let’s say it’s content that I generated. Is it still solely my speech if Meta spends a fortune to amplify it to so many users who wouldn’t have heard it, but for Meta making that decision, or is it now both of our speech? I’m not sure exactly where that line is. If it’s obviously purely Meta content, then it’s pretty simple, but there are going to be a lot of cases in between. But I think the broader question is: is 230 in its current form still meeting its purpose? I don’t think it is, but I know there’s no consensus over what that ought to look like. And that is part of what has hampered us from doing anything. But even the smallish stuff, the small-bore stuff like don’t allow your platforms to encourage people to commit suicide has been hard for us to do in the social media context. Really, litigation has been more effective than legislation because we haven’t been able to legislate. If you look now to AI, we made some progress legislatively, at least in terms of the process in trying to prevent AI chatbots from telling kids how to commit suicide or keep it from their parents. But even those little things are still not at the finish line. So what does that say about our capacity to do the big things? Historically, when stuff is tough like this, we wait for calamity. And in the case of AI, because that calamity could be just massive, we can’t afford to do that. It does seem to me like we’re about to run a fairly large-scale A/B test on the value of the 230 immunity. Whatever ChatGPT is saying to people is very clearly OpenAI speech. And then somewhere next to that is the kind of clarification law that AI companies are talking about in terms of agents and things like the Hugging Face attack, where they don’t know if a user directing an agent to do something bad is their fault. Whether that liability should pass through them if I’m the person who told my agent to go do some horrible hacking attack. I talked to the tech industry, and the thing they want from legislation is actual clarity in the law. And instead, what we’ve gotten is this non-binding agreement with the Trump administration that simply says things like, “The companies will empower an internal team to ensure all of the controls, monitoring, and detection are operating as intended,” which feels like nothing. To me, that is literally meaningless. Does that strike you as toothless and as meaningless as it does to me? Yeah, it certainly does. I think it really is meaningless. I’ve been thinking about the question you just raised. It’s interesting that you brought that up, and that is: where does the legal liability rest today? So if you have these bots hacking into Australian websites, or you have them hacking into Hugging Face, clearly that is the responsibility of these models because that’s not being done at the behest of one of their customers. But let’s say you build a tool for the purpose of hacking, or that can be used for hacking, and one of your clients, one of your users, uses that tool that way. How much of that is your responsibility? Is it analogous to somebody who sells a gun and then the gun seller says, “Well, it’s not my responsibility for how they go out and use it”? Or is it someone who sells a weapon that is flawed and causes injury because of a flaw in the manufacture of the weapon? Or is it selling a weapon to a kid who’s not eligible to buy it, or to a convict, or not using due diligence? What is AI responsibility? What is the company’s responsibility in the sophistication of the products if they’re building things that the plain use can be to commit ransomware attacks? I think that the gun metaphor is a good one. It’s easy to grab onto for a variety of reasons, but I tend to think that it kind of obscures the reality of the capabilities of the model. So OpenAI’s latest model, Mythos, can be used for offensive cyberattacks, and it can be used on defense too. And the cybersecurity firms that we cover, they want it. Their argument is, “We actually need to take humans out of the loop so that our cyber defenses can be as responsive and fast to incoming AI-powered cyberattacks. And if we insistently put humans in the loop, then we will be slow, and the attackers won’t have the same morals or ethical qualms we do, and we’ll lose.” And this is all just a dance around, well, the model in this case can be used on offense and on defense, and it should be equally capable on both sides. I don’t see a regulatory framework that provides clarity for any of the frontier labs to be able to sell that thing and not feel like, well, if the equally capable defensive model gets used in the wrong way, that they won’t be liable. And I think that’s frozen some of this development in its tracks. I’m wondering, again, just from a legislative perspective, these are thorny questions. If you try to write a bill to provide some of this clarity, it will necessarily have to be vague on the margins because the technology is moving so fast. What is that debate like for you? Well, I think what it means is that this debate is going to keep evolving as AI evolves and at a speed that is really much faster than Congress has proven capable of legislating. I like to say that in the technology world, Moore’s law is increasing productivity, doubling it every 18 months, and with AI, maybe that’s quadrupling it every five months. The legislative calendar works more on a Murphy’s law timetable. And I think that disparity is even more pronounced in the area of AI. What I think that translates into is that we’re going to need a regulatory body that has real discretion and real expertise that can move a lot quicker than Congress can. Whether that looks like an FDA for artificial intelligence or some other agency, I think that’s ultimately going to be necessary. That agency’s going to have to have real power and teeth and real expertise. It’s going to be a challenge, as with these independent monitors: if there’s a government component there, how do we afford that? Because there’s so much talent, and that talent can draw much bigger salaries in the private sector. But I do think we’re going to need a regulatory body in addition to whatever the overarching statutory framework is. Well, in addition to being a bad lawyer, I was a very bad law student, mostly majored in Miller Lite, but it’s fine because every foundational precedent that I was taught in law school in 2003 has been overturned by this Supreme Court. And the most important one in this context is Chevron deference. In Loper Bright, the Court said we no longer have to defer to the experts at agencies; go ahead, corporate interests, tie every regulation up in litigation, and then we’ll decide. The Roberts Court will decide if these are appropriate regulations. This strikes me as a huge problem for the idea that we will set up a regulatory body to oversee AI. Because what you’d want is a bunch of experts in that agency making decisions on the margin, saying, “Okay, here’s a vague enabling statute that Congress passed that said, ‘Pass all appropriate regulations to keep us safe.’” And then you would write a bunch of regulations inside the agency, and until last year — no, I’m sorry, until 2024 — that’s how it would work. That’s gone. If you enable an agency in that way, OpenAI can show up and say, “That’s inappropriate. We actually don’t think that’s right.” And they would tie it up in litigation forever. How do you get over that if you want to set up a regulatory body? It’s very difficult for the reason you mentioned. The Supreme Court is a lot of things today. It is now a partisan institution. It is, in my view, a right-wing social institution, but there’s one common denominator that runs through its decisions. That common denominator is not always pro-Trump. The common denominator is money. That court makes the decision where the money is. So when it involves, for example, tariff decisions, it will rule against the president because tariffs impact the money. It will allow the president to remove independent heads of agencies, except maybe the Federal Reserve, because that’s where the money is. And so you’re right, they are corporatists first and foremost, and they’re more than content to take power away from Congress to regulate and away from regulatory bodies. It does mean that we are going to have to draft legislation in this area empowering the agency and making it as specific as we can, make these agencies as powerful as we can, delegate as well as we can, knowing that the Supreme Court will strike down anything that is in any way ambiguous if it suits the money interests. There’s a lot of political realignment happening around AI and with privacy, which I want to talk about in a little bit. It feels like there’s more across-the-board antipathy towards AI companies that might be harnessed into writing an actual piece of legislation and getting it passed. We see it on the right and the left. You see it in the backlash to data centers. This is a story that is happening on the right and the left. We have a big story about the Utah data center: Kevin O’Leary got himself in trouble for calling all of his detractors Chinese agents, and he had to issue an apology because the people of Utah are not Chinese agents. That doesn’t feel like a left-right breakdown to me. How does that feel to you? Is this something that you can actually channel from across the spectrum into a meaningful policy outcome? I don’t think this is a partisan issue. I mean, it’s certainly true that the president is all in for AI, no restrictions, no impediments, no pacing, no nothing. And there are some of my Republican colleagues here that are going along with that agenda. There are others that are hearing from their constituents who don’t want a data center in their backyard or are concerned they’re going to lose their job as a result of AI or their life as a result of AI. And I think we are seeing the kind of coalition where you have populists on the left and populists on the right coming together, initially around concerns over jobs and job losses, and what it will mean for working families. But now, across a broader set of issues, the data centers are the most tangible way for people to express their concerns about their neighborhood, but also about their job, about the technology, and about the uncertainty of it all. And I think the initial concern, and I still think it’s quite a dominant concern, is over the impact on jobs and the nature of work. But what has reached the fore more recently is concern about runaway AI and the danger it poses. What will the concern look like a month or two months from now? I don’t know, but I do think it’s not going away. And as the AI bills — and I’ve introduced several — indicate, not just mine, but others, most of those bills are bipartisan. Some around the data centers are predominantly democratic bills. My bill on the data centers would force them to pay the costs so that consumers don’t have to. It’s not the voluntary, we pledge, we might, we’d like to; just don’t provide anything that will force us to observe those costs. Those are mostly democratic bills. Republicans have largely adhered to the president’s line about not posing any real imposition. They want to pass fig-leaf kind of stuff, but not real stuff. But I think increasingly, as people hear from their constituents, there’ll be bipartisan interest in legislating this area. You’re on the antitrust committee. One of the more surprising bits of the structural realignment that we see around AI is the notion that there should be an antitrust exemption so these companies can coordinate on cybersecurity. Lina Khan, who is Biden’s FTC chair, said, “You don’t need an antitrust exemption.” We just had Jonathan Kanter on the show, who ran antitrust at the DOJ under Biden. He was like, “You don’t need an antitrust exemption for this.” But at the same time, you have David Sacks agreeing with Lina Khan, which is not a thing that happened during the Biden administration. You have a lot of what you would call the little tech players in the tech right being like, “No, no, no, this is how you’re going to make a cartel. And we didn’t like the Google and Apple cartel that kept a lot of our investments from going. They have to compete in the marketplace.” That’s surprising to me. You don’t often see David Sacks and Lina Khan on the same side of an issue. You’re on the antitrust committee. Do you think the companies need an antitrust exemption? I think a very narrow antitrust exemption could be very helpful in addressing foreign distillation efforts. We would want companies to be able to talk to each other to address a mass distillation effort