Working at the intersection of AI and law

Does AI policy actually need lawyers? For what kinds of issues? And if it does, is law school worth three years and the price tag? We get questions like these all the time, from students and technologists considering a degree to lawyers considering a pivot into policy.

Kevin Frazier is well-equipped to weigh in on these questions, which is why I was excited to interview him. He’s the inaugural AI innovation and law fellow at the University of Texas School of Law, where he founded the AI innovation and law program. He also works with the Abundance Institute and hosts the Scaling Laws podcast at Lawfare. His own route took him from the Oregon governor’s office to Google to clerking in the Montana state supreme court to teaching constitutional law (and much in between), giving him a broad read on the intersection of law, government, and AI.

In our conversation, we cover:

• What lawyers can do in AI policy that other people can’t

• Whether to go to law school, and what to do once you’re there

• Which AI policy questions Kevin most wants lawyers to take on

• What it’s like to change your mind and stances on issues publicly

Remco Zwetsloot: You run the AI innovation and law program at UT, you’re engaged through the Abundance Institute, and you host a podcast. How do you split your time? What does your work look like, and how has it evolved?

Kevin Frazier: My role is unique. I’m the inaugural AI innovation and law fellow at the University of Texas School of Law, where I started our AI innovation and law program. Our goal there is to think about two things.

The first is AI for law: how the next generation of lawyers will learn to use AI tools in legal practice. That part of the program helps students learn to use Harvey, Lora, Claude for law, and all of these new tools. Then we can stop reading headlines about an attorney submitting a brief with hallucinated citations—I’ll save that rant for another time.

The part of the program I’m most focused on is the law of AI. How do we think about novel regulatory regimes, and really the future of governance, now that we have this incredible emerging technology? In the fall I teach an AI and law survey course. Over 14 weeks we run through the fundamentals of AI, everything from what a neural network is, to backpropagation, to what a world model is, to AI and consumer protection, AI and free speech, AI and intellectual property. In the spring I teach an AI and law policy workshop, where we work with organizations from Meta to Engine to Meridian International Group to identify policy and legal questions that have yet to be answered, and have students at UT help fill that gap.

I’m also the host of the Scaling Laws podcast, which aims to help people get a strong understanding of the policy ramifications of AI, where that discourse is headed, and perhaps where it should be headed, all grounded in a shared interest in harnessing AI for good while mitigating its risks (to sound like a broken record, like everyone else does).

Then with the Abundance Institute, that’s where I get to apply the things I study at UT and translate them into a more specific policy role at both the state and federal level. That means helping lawmakers understand the ramifications of this technology and get a sense of what a proper regulatory regime would look like, again for that same mission of making sure AI goes well. In that capacity it looks like testimony before Congress, going to state capitals and sharing my two cents with lawmakers, and engaging with stakeholders, from academics to think tank folks, to get a sense of what a good regulatory approach to this technology might be.

How did you end up developing this portfolio? And for people who might want to do something similar, how did it actually happen?

Kevin Frazier: My first job out of college was working for the governor of Oregon, where I got to see how the sausage is made. As people may have surmised, it is sometimes a pretty gnarly process, and not always awe-inspiring in terms of efficiency. It does not always live up to the standards of the West Wing.

So I thought, I want to see the exact opposite of this. That’s when I went and worked for big tech. I worked as a legal analyst at Google and saw what it’s like for an organization to operate at a global scale and with incredible speed, able to change policy literally at the click of a button and suddenly affect billions of lives. I realized there’s quite a mismatch going on. We have this incredible promise of technology that sometimes goes well and sometimes goes really wrong, and we have government workers who are well-intentioned and want to do the right thing but face their own set of constraints.

We have this incredible promise of technology that sometimes goes well and sometimes goes really wrong, and we have government workers who are well-intentioned and want to do the right thing but face their own set of constraints.

From those two experiences, I’ve tried to champion the idea of being a translator between those communities, and helping make sure we have institutional capacity at the government level that’s ready to assist with innovation and guide it for the benefit and general welfare of the public. That’s an abstract mission, and fortunately it’s led me in all sorts of directions.

I’ve found really strange ways to contribute and leaned into them anytime they felt inspiring and like an opportunity to learn. By virtue of being a really curious person, and someone who admittedly can get bored quickly, I found myself in a lot of interesting rooms. I was a clerk on the Montana Supreme Court, where I got to see, at the highest level of state government, what the institutional capacity of that institution is, what questions are being raised, and how the law interacts with new technology.

That led to a research fellowship with the Institute for Law and AI, where I worked with Christoph Winter and the team studying nascent legal questions posed by AI.

That then led to a role at St. Thomas University College of Law, where I taught constitutional law, civil procedure, and administrative law. While there I got pulled into Lawfare, a publication that covers national security, emerging technology, and the rule of law. I carved out an AI beat there and got to keep being really curious, ask really dumb questions, meet fascinating people, and share their insights mixed with what I hope were some of my own novel contributions.

Then Bobby Chesney, the dean here at UT Law, must have been doing his dishes or taking out the trash when he heard me on a Lawfare podcast talking about AI, and thought, this guy sounds somewhat competent, and I want to start an AI innovation and law program. I got to move to Austin and help start the program, and then, again by being willing to share my two cents and dive into conversations, I met the great folks at the Abundance Institute.1 Both UT and Abundance care a lot about the nexus of law and AI, so we came to a good arrangement where I can do everything we need to do at the law school to prepare students to contribute to this question, and then share that knowledge back out through the forums Abundance makes possible.

Trying to distill the themes there: it won’t look the same for everyone, but incredible curiosity, following your nose, trying to be at an intersection that’s pretty nascent, and carving out a little niche that ends up having much broader relevance than you’d expect. Does that seem right?

Kevin Frazier: That’s right. Curiosity, a willingness to ask questions, and then the courage, or stupidity in some cases, to pursue the answers. You have to know you’ll sometimes get the answer wrong, or hold a different stance six months later. By virtue of being excited about starting conversations, I have more skeletons in my closet than a cemetery. You can find op-eds I wrote in high school that I refuse to read. But there’s real value in being willing to share your perspective even when you might be wrong, and to learn from others.

So I’d encourage people not to be afraid to admit when they’re bored, and to go seek out those they want to learn from. As long as you’re applying real enthusiasm to a question, it’s going to lead to good things.

I’d encourage people not to be afraid to admit when they’re bored, and to go seek out those they want to learn from.

I dropped out of a PhD after souring on academia. A big reason was that the incentives for someone in that career really aren’t pointed in this direction (we’ll do a whole separate post on this), but you’ve come to a good arrangement. Why do you think this approach of policy engagement isn’t more common in academia, especially post-tenure, and what would you like to see academics do more of?

Kevin Frazier: First, I’m fortunate to be in a place where I’ve been able to take a lot of risks and execute on a lot of gambles, and that’s not the case for everyone, depending on their familial or financial situation.

I’d really encourage academics to embrace what I think is the purpose of academia: to share expertise and help society answer important questions. And that isn’t always incentivized by tenure.

The core of it is acknowledging that there’s a real shortage of good answers to really basic questions. There are folks on the Hill and in capitols around the country who just need a two-page explainer on reinforcement learning from AI feedback. And yet, where is it? We have hundreds of computer scientists in PhD positions and tons of academics who study emerging technology, while policymakers and civil society are screaming that they don’t know what to think of this new AI thing, four years in.

So what I’d encourage academics to do is the equivalent of touch grass. Go outside, get off your campus, go talk to normal people, and in particular go talk to policymakers about the questions they have. Then regard it as your mission to help answer those questions. Suddenly your research agenda becomes a lot clearer, because you’ll know exactly what you need to dive into.

What I’d encourage academics to do is the equivalent of touch grass. Go outside, get off your campus, go talk to normal people, and in particular go talk to policymakers about the questions they have.

If a legal colleague, or even a law student, came to you wanting to work on AI policy, what do you think they can contribute with their legal training that someone else might struggle to?

Kevin Frazier: You can take, for example, the Trump administration’s recent executive order on AI, which created a voluntary vetting process for frontier AI models. We could analyze that from a technical standpoint, a policy standpoint, or the political science of it, but ultimately you need lawyers to assess whether it’s even legal, and if so, why, and if not, why not. Those are questions only lawyers can answer.

The basis on which any policy is grounded always needs to be resolved, given that we have a government of limited, enumerated powers, something every student learns in constitutional law. It’s our job to assess the extent to which that’s true. If there isn’t a stable legal basis for a policy idea, then it’s not good policy. So lawyers are a fundamental part of any policy development cycle. That’s a really exciting opportunity for law students and lawyers, because regardless of how you think about AI or its trajectory, there’s going to be a severe need for people who can analyze laws written decades, if not centuries, ago and figure out how they map onto novel technology.

If you could wave a magic wand and direct a bunch of people’s effort, what are your top two or three specific AI questions you’d love lawyers to dig into more?

Kevin Frazier: First, borrowing from Professor Ryan Calo at the University of Washington School of Law: we need to critically analyze and update our entire perspective on privacy law. If you want better AI, you need better data. Privacy laws, very well-intentioned and often written in the 1970s, are not grounded in the idea that we should share high-quality data in sensitive domains. So how do we update privacy laws to make sure we can build the most transformative aspects of AI, like AI for health care, AI for education, and AI for people in unique and sensitive circumstances? That’s one.

Another, and I hope folks will get in touch if they’re fired up about it, is constitutional AI. I’m fascinated that we’ve seen from Anthropic and others that devising a set of values and principles, then training models on them, appears to be a really promising alignment technique. And yet we haven’t developed governance infrastructure for how to think through who should set those values, when they should change, at what pace, for what reasons, and by what threshold. Those are exactly the questions I ask about constitutions and constitution-making. So this is an exciting moment for people who have been nerding out about con law and reading treatises and the Federalist Papers to get involved in some weighty, interesting conversations.

You’ve also written about how lawyers erected much of the bureaucratic apparatus that has prevented us from building housing, completing transit projects, and otherwise responding to 21st-century concerns. What do you think lawyers get most wrong when they approach AI policy, or tech policy more broadly?

Kevin Frazier: Lawyers are not known for their creativity, and that’s something I try to address at UT. We’re not known for imagination either. And yet this is exactly a period where a technology is moving at an exponential rate and being applied in such new and exciting circumstances that failing to apply creativity and imagination to any law right now is likely to produce a bad law.

We have whole law review articles about “zombie laws”. Zombie laws are the product of the fact that once a law gets on the books, it’s really hard to get it off. This is why I can’t buy beer in Austin before 10 a.m. on a Sunday, so do not invite me to your Super Bowl party, because I’ll forget to buy beer.

Why do we have bad policy? Usually because laws aren’t dynamic, aren’t designed for an evolving, emerging technology. This is where we need lawyers to lean into concepts like sunset clauses, so laws don’t become permanent and can evolve as the technology changes. It’s why lawyers need to recognize the value of retrospective review, which is really just checking in every once in a while to ask whether a law is working as intended. And it’s why we need lawyers to talk with and learn from computer scientists, so they write laws that don’t, for example, cover technology they shouldn’t. Some proposed definitions of AI are so broad they’d regulate a calculator. There needs to be a lot of epistemic humility baked into writing laws in this domain.

Horizon advises hundreds of people a year who are wondering whether to go into law at all. Let me play devil’s advocate: there are part-time and evening masters programs in DC compatible with a full-time job, so you can be in the field right away while still learning. Why should someone go to law school instead, and what kind of person would you tell to do the masters?

Kevin Frazier: For folks who are law-curious, if you want to go that route, you have to find a program that’s aware the practice of law will not look the same in two years, five years, and certainly not in ten. First and foremost, make sure you’re considering a program that will prepare you to be a lawyer in the age of AI, which isn’t true of all law schools. I like to think it’s true here at UT. That’s a threshold question. If you go to a website and see no mention of AI, or the only mentions are “you can’t use AI for this” and “you can’t use AI for that,” that’s probably not a school you want to spend your time in.

The second question is whether you’re excited about questions only lawyers can answer. Go see: can I go to ChatGPT and draft a will? Yes, and it’ll probably be pretty good. Can I go into Harvey and draft a complaint? Yes, and it’ll be pretty good. If you’re going to law school to do the things AI already does well, there will be fewer and fewer of those specific jobs.

If instead you’re excited about learning to think like a lawyer, and about what I’ll call the legal scaffolding of society, that’s the case for going. If you can’t wait to think about regulatory design, the future of constitutional law, and what I’d call the future of governance studies. Those are heavy topics a good law school will teach you to explore, and they’re hard to develop without the immersive experience law school provides. So if training your mind to think differently and answer questions others aren’t prepared to answer really excites you, that’s why you should go to law school. If you just want a JD next to your name, it’s probably not the best time to go.

Suppose that pitch is persuasive and someone is a 1L planning how to spend their time. What can people do while studying? Are there extracurriculars you recommend? How should they spend their summers?

Kevin Frazier: First, get involved with any AI initiatives on campus, which may be at the law school, but your better bet is to go hang out with the computer science kids. It’s easier for a computer science student to become a lawyer than for a pre-law or JD student to develop a good understanding of computer science. Go get to know the CS professor who also writes on AI policy, or the CS students doing clubs and hackathons, and understand the real technical aspects of AI. That gives you a big advantage when you’re looking for unique roles at the nexus of AI and law.

Second, make full use of your elective classes to learn those more technical aspects of AI and adjacent policy questions. Some people feel they need to use every elective for a bar prep course. I get that strategy, but life is short and law school is even shorter. Make the most of it by learning things that excite you and set you up for professional success, not just the rules of evidence, which you’ll get down in a week or two.

Life is short and law school is even shorter. Make the most of it by learning things that excite you and set you up for professional success, not just the rules of evidence, which you’ll get down in a week or two.

And then, start writing. Start asking big questions, diving into them, and sharing your thinking. The more you show you’re a curious person actively thinking about how the law applies to AI, the more interesting people you’ll meet and the more interesting spaces you’ll be invited into.

Are there particular institutions, events, or resources you find yourself recommending to students at the intersection of law and AI?

Kevin Frazier: I’ll give the standard law professor answer, which is that it depends on the campus. Listen to Scaling Laws as a good place to start, because I’m a shameless academic. If you hear a guest you like, or an AI law professor on another podcast, check out their work. Follow people like Peter Salib at Houston, Yonathan Arbel at Alabama, and Alan Rozenshtein at Minnesota. Get a sense of what they’re writing and see if it’s something you want to explore. Basically, fake it until you make it. Find whoever you think is doing good work, find out what they’re up to on campus, go hang out with them, and you’ll end up in a good spot.

Are there ways you’d recommend people spend their summers? It depends, of course, but is there a common pattern you see across the people who get the most out of theirs?

For summers, be creative. These are the last summers of your life, so, not to be too depressing, be wild, be crazy, and go find the thing no one else is thinking about. If I were a rising 2L right now, I’d email every startup mentioned in The Information last week and say, “Hey, I’m a law student, I’m super AI-curious, I think I could add value, and I’m willing to do any legal or policy research you need. Let me know when I can show up.” I bet you’d get at least two responses. Maybe not a ton, but somebody is going to love the gumption of a person willing to show up and do anything.

Or go find a congressional internship, a state legislator, or a state agency. There’s incredible work being done in North Dakota on integrating AI. The North Dakota Office of Legislative Council is one of the earlier adopters of AI tools in the legislative process. There are agencies up and down California doing incredible stuff with AI. Go find the people on the vanguard of AI legal questions, and you’ll end up in a good spot.

Changing track from the pure law path: you wrote a great piece, “True Confessions of an AI Flip-Flopper,” about how your views have shifted. I’ve been in the field since 2018 and still don’t know what I think about most questions. It’s rare for people to change their minds, and even fewer people do that publicly. What was it like to write that piece, and what reactions did you get, especially as the field becomes more tribal?

It’s really important to acknowledge what your fundamental views and principles are, keep those close to heart, and then give yourself the freedom to acknowledge that their application may change over time, and that’s okay. It’s important to share that so other folks feel willing to do it too. For us to get AI policy right, whatever that means, we’re going to need to stop relying on dogma, rise above partisan politics, and be really nimble in how we respond to new advances in the technology.

For us to get AI policy right, whatever that means, we’re going to need to stop relying on dogma, rise above partisan politics, and be really nimble in how we respond to new advances in the technology.

So writing that piece was my small contribution to say, hey, folks, these are wild times, and we need to be nimble and epistemically humble so we can truly harness the benefits of AI and mitigate its risks. Those benefits are going to change, and so are those risks, and the best response will be even harder to predict from one point to the next. The more you state your principles and identify your values, the easier it is to admit you need to change course because you’ve seen new evidence. That, to me, is the sign of a good lawyer and a good policy practitioner.

You’ve said that you don’t know yet whether there will be a follow-on post. Are there things you’ve been tracking that recently feel different?

I wouldn’t be surprised if a piece is coming, because the gulf between frontier AI and what I’ll call boring AI is really growing. Boring AI is, “I used AI to generate a brief,” or “I used AI to detect a pattern in medical data I hadn’t seen before.” Those are transformative use cases, but technically they’re wildly boring, and that kind of AI is only going to get better. At the frontier, when you look at what the UK AI Security Institute saw from ChatGPT 5.5 and from Mythos, and extrapolate based on scaling laws to what we may see in cyber and bio capabilities, you’re looking at something else entirely.

Having laws map onto both kinds of AI is really hard, and maybe shouldn’t even be pursued. There’s a temptation to write one rule to govern them all, and that’s not how this technology, or our use of it, is evolving. For most people, in most contexts, and arguably for some of the most transformative use cases, those boring AI tools might be all they need to change the world. Who needs and who can leverage the frontier models is a very different question.

I don’t think a lawyer needs Mythos to draft a compelling brief. We’ve already seen lawyers do incredible work with less sophisticated tools. Is it controversial to say maybe lawyers don’t need the latest and greatest AI? Maybe. But it’s a statement I’m willing to live with, and it’s become more true for me in this post-Mythos moment. I’m not sure it’s true for everyone yet, so we’ll see whether it rises to the level of a follow-on post.

Any final thoughts, a piece of advice or encouragement you’d want to leave a reader with?

The biggest piece of advice would be: don’t succumb to patterns. Don’t feel you need to follow the traditional path in an era where everything is being upended. I think the riskiest thing to do is to follow the well-trodden path, and the safest approach is to be wild and wildly curious.

I think the riskiest thing to do is to follow the well-trodden path, and the safest approach is to be wild and wildly curious.

That would be my biggest piece of advice. And for folks who want to pursue crazy ideas, feel free to send me an email. Happy to chat.

Big thanks to Kevin for taking the time. If you enjoyed this and want to learn more:

• Kevin on Lawfare

• Kevin’s Substack, Appleseed AI

• Our guides to law school and other graduate school for policy work on emergingtechpolicy.org

• The AI Innovation and Law Program at Texas Law, which Kevin leads