What could AI guardrails look like? Northwestern expert explains

Abby kind of just ran us through some of the nuts and bolts about the warnings over AI we've been hearing recently. Some of those warnings are turning into calls for action in Washington. President Trump, though, is kind of pushing back on this. Yeah, saying that we don't need guardrails in that sense. So we wanted to take a deeper dive into what government intervention on AI would look like, what it could look like, and this morning we're talking more about this with Daniel Linna, professor at Northwestern's Pritzker School of Law and McCormick School of Engineering Thank you so much for joining us. Yeah, glad to be here. So when lawmakers are talking about putting guardrails on AI, I mean, what does that actually look like, do you think? Well, I think the context is really important here, and sometimes the discussion seems to suggest that it's the Wild West and there are no laws or regulation that apply to AI. But of course, there's plenty of existing law that applies to AI use cases. Now there's a couple of gaps there, and states like Illinois and others are addressing that. And even the federal government is acting in different ways. There was an executive order a few months ago from the Trump administration in an agreement with the Frontier Labs to get a look at models before they released to the public. So the question, one of the things that's really in the top of everyone's mind is this discussion about an existential threat to humanity. And while there are risks for AI, there's really not evidence to support this idea of an existential threat. And one of the things driving this is this hugging face incident with open AI. But if you dig into that a little bit, it's not really losing control of AI as much as it was a test inside of open AI, where they took away the guardrails from their AI tools, the guardrails that are on the tools we commonly use every day. And then those tools ended up hacking into hugging face, right? So that's important context. Yeah, absolutely. Well, thank you for clearing that up. So you mentioned gaps, all right? So let's zero in on what you think those gaps are. They are regulated in some sense, but does more need to be done? Well, yeah, I think that one of the major gaps is something that recent AI legislation in Illinois attempted to address. And that is the safety law that would go into effect starting in January 2027, requiring risk assessment frameworks and public disclosure of them, independent compliance audits, and more transparency. And that was one of the problems around this hugging face incident, is that it took a long time for the facts to come out. And we really need to be in a regime where when something like that happens, there's a requirement to disclosure. And we've got requirements like that for breaches, data breaches, things like that. So there could definitely be some frameworks in place, and that would be really helpful. That's just one example. There are some questions about the liability for individuals using these tools. Courts are sorting that out to a great extent. But there's some uncertainty there that legislatures may want to step in and address as well. Daniel, I do have a question about how this thing came out with Anthropic last week. And then all of these AI CEOs, they came out and were immediately like, yes, please regulate us. We need regulation. But these are CEOs of companies, Daniel. Why aren't they just regulating themselves? Why are they asking the government to do it for them, do you think, instead of just saying like, okay, as a company, we're gonna leave the guardrails because we saw the hugging face? Do you know what I'm asking? I don't know if I'm- Yes, I know exactly what you're asking. Okay, okay. Yeah, cuz it's weird. It's difficult to understand. It's difficult to understand the motivations. Giving them the benefit of the doubt, I believe that the people who are working on these tools do have concerns about that. These tools are quite capable now, even though I don't think we're anywhere near artificial general intelligence. But nonetheless, they're capable and they can cause harm. So there's a genuine interest in the industry acting responsibly and having reasonable regulation in place. Now, the other concern right now is this issue of regulatory capture. And of course, these frontier labs now, they're spending billions of dollars and they're being challenged by open source models that don't have nearly the resource. So there's a lot of skepticism that this is a bit of a play for regulatory capture. This discussion about the public and governments having an interest, I mean, that's not usually the way the free markets operate, right? So there's a lot of skepticism about this being a play to kind of fortify the incumbents in the marketplace and prevent competition. Interesting. Right, right, yeah, exactly. They're doing it in a way in their own interest. Now, I wanna get your reaction to President Trump's comments on this. He pointed out that our main competitor would be China in terms of developing this technology. And that we don't wanna let up an inch when it comes to that. We wanna be on the forefront. America wants to be on the forefront of this. How do you respond to that? Well, I think there's two important things to consider there. I mean, national security is absolutely a very important consideration. We're already seeing how artificial intelligence is being used in national security around the globe. So absolutely, the US needs to be a leader in this space. At the same time, there's this false dichotomy sometimes about innovation, where if we regulate, we're gonna kill innovation. So we have to be very smart about the way that we regulate. I don't think anyone would say we should totally deregulate air travel in the US, for example. However, if the regulation is burdensome and wrongheaded, that can create a lot of unnecessary costs, right? So the context really matters. What harms are we really looking at? I think all too often, we're not talking about particularized harms that we're trying to address, right, thinking about that. But absolutely, the national security implications of AI have to be considered in making these decisions. I am curious. I know that we're talking about Congress. They wanna come in. They wanna have these hearings and regulate technology. And I was thinking the other day about the 2023 TikTok hearings, where lawmakers asked them, it kind of became a little bit of a social media joke, some of the questions we were hearing. And do you think Congress is equipped to regulate technology that is changing so quickly? I know that I don't know about, I'm sure that a lot of people in Congress, they're not tech experts. So what do you do in that regard? Yeah, I think that there have been public displays that show that elected officials have a lot to learn about AI. At the same time, I've heard talks, for example, I was at Notre Dame for an event like 18 months ago, and Senator Todd Young was there. And I was very impressed about what he had to say about AI. And so they're surrounding themselves with knowledgeable people. They have to do better in this respect. I think this goes to another thing. I work a lot with judges and courts, and we need to make sure that people in government have the right tools available to them, but the right training and the other professionals surrounding them to understand AI, to understand how they can use it responsibly and well in what they're doing, and to understand what the real risks are. Understand how to be able to ask questions when these tech CEOs appear in front of you. And more and more, that's a competence that all of us in society need to have. I mean, AI is not just one general monolithic tool. This pause, for example, we're talking about pausing AI, but what does that mean? Which tools? Are we gonna say the protein folding tools that are creating breakthroughs? We're gonna pause that as well, right? So there's a lot of details to work out here. And absolutely, we need our legislators to be more knowledgeable and surround themselves with experts. That's a good point. Very well said. Daniel Linna, thank you so much for your perspective. I'm reassured by your tone. I am too. Because it's not like, oh my gosh. It's not the sky is falling. Exactly. Thank you, sir. It feels, yeah. Thank you, sir. Because sometimes it can feel like that. Wonderful to join you. Thank you. Yeah, that was really nice. And here's the thing. All these older judges, these aging congressmen, all they have to do is ask Claude, get me up to speed on AI. Well, they say, I mean, he mentioned it, it's who you surround yourself with. No, absolutely. Sometimes it's not about the person. I'm kidding, but yeah. It's who you're surrounding yourself with. So it's true. You need to seek out the experts in that regard. You do.