AI and the DMZ

What I learned at SAIPCON 2026

Every terrific conference is terrific in its own way, especially when there’s recognition of the risks (if not the terror) of not getting it right. SAIPCON 2026 “Frontier Issues in AI Governance: Agents at the Gate” in Seoul, South Korea on June 16-17 was committed to addressing every pressing question imaginable about AI and the agents at the gate. Institutions, nations, and disciplines are under immense pressure, both technological and political, to act, to implement and revise domestic and international policies. The speakers, from across the globe, were blunt and reflective about the risks, costs, and opportunities of AI in its many forms. Nations do not want to feel dependent but there are barriers to cooperation. Nobody wants to feel like they are lagging behind, except those who withdraw as a deliberate strategy.

And yet there was a marked absence of anxiety in the packed Plaza Hotel ballroom for these two days. The tech-savvy audience of academics and global leaders were all fully attentive, taking notes, impressed by and grateful for the simultaneous translation technology (you could listen, see it on screen, or have it sent to your devices). There’s so much fear and fear-mongering in the AI discourse that its absence in the room was noticeable, even during the panels on sovereignty, self-sufficiency, and defense technology. I was struck too by the candor about the reality of U.S. and Chinese dominance and what it means for countries like Korea, Japan, and Vietnam, and for the EU.

The conference was organized by Yong Lim, a law professor at Seoul National University and head of the AI Policy Initiative, and Eun-Ju Lee, professor of communication at SNU; co-hosted by Christopher Yoo, professor of law at Penn and Founding Director of the Center for Technology, Innovation & Competition, and Haksoo Ko of SNU. Yong Lim and the AI Policy Initiative have been hosting these conferences since 2017 and this one, everyone said, was the biggest and most well attended yet.

There were 15 sessions in total, from AI and law, AI and health care, AI and national security, AI and energy, to AI and education and much in between (list below). Everyone could follow along in English or Korean or both. For some panels, slides did much of the work. I took notes on almost every panel. Anyone focused on AI and higher education, as I have been for the past two years, should pay attention to what’s happening in other fields, since that’s where our graduates will be working.

I was invited by Gus Hurwitz, a senior fellow at Penn and academic director of the Center for Technology, Innovation, and Competition. I’d met Gus last fall through the excellent Sigal Ben-Porath, when I spoke virtually to a group of graduate students under Penn’s Public Interest Technology program about their use of AI. The panel Gus organized for SAIPCON, “AI in the Ivory Tower: Human Agency in Education and Research,” kicked things off and while I did not have colorful slides I made a point that I’ve found myself making a lot these days, which is how much fictional representations of AI — popular movies (like The Matrix), so much sci fi, Star Trek — dominate public understanding (and misunderstanding) of AI. Fiction and poetry are centuries ahead of LLMs in creating artificial beings and worlds that readers believe are real and will defend, not to mention provoke wars. Poets know that language is dangerous. What can we bring to the discussion of AI fear or AI desire? I ended on the Korean poet Kim Sowǒl’s heartbreaking poem “Azaleas” (1925).

Gus Hurwitz, me, Kim Sowol's poem "Azaleas"

Gus’s concern (he shared with the panel an important piece he’s working on about AI in education) is AI, equity, and relationships — the importance of tacit knowledge beyond what AI knows, which can only be transferred through relationships. Universities have become “the same” (a point I share) and AI may bring us back to the individual.

It was great to go first and then just listen. Here are my notes from some of the best talks and panels:

A Tale of Two AIs: Augmentation and Automation in Central Banking

Fernando Perez-Cruz (Bank for International Settlements) introduced a theme that carried through several panels: separating chatbots as assistants from LLMs as enablers of task-by-task automation, where you know what the task is and what data should be used. But, the Karpathy retweet: “You can outsource your thinking, but you can’t outsource your understanding.” The big question: is there a link between productivity and tokens used?

The panel discussion, Beyond the Chatbot: Promises, Limitations and the Risk Landscape of AI in Finance featured the brilliant Soohyung Lee and Trisha Ray.

AI, Sovereignty & National Strategy

This was a “fireside chat” between Yong Lim and Gregory Allen about AI, semiconductor strategy, and what national sovereignty actually means. We had all just lost access to Fable a few days before (which Gus and I were mourning) so sovereignty was on everyone’s minds. I was listening so hard I didn’t take notes.

AI and Legal Tech

An excellent panel largely about the “human-in-command” design: agents should never ever send emails to a client, an opponent, or a court. Discussions of LLM risk in workflows (judgment summarization, precedent retrieval, legal issue extraction, evidence analysis). I was there mostly to learn from Trevor Quick, who reported that nearly 69% of legal professionals now use some form of GenAI (up from 31% in 2025 and 27% in 2024). Others talked about the market for legal AI beyond Harvey and governance and policy questions. Plus, of course, how to train future lawyers….

AI, Competition & Industrial Policy

A fascinating conversation with Ariel Ezrachi, Friso Bostoen, John Newman, and others about antitrust law and what happens when the technology is moving fast but the law is slow. “Should competition regimes move faster?” Newman asked. Yes, because delay has costs.

AI Transformation and Regulation

Christopher Yoo moderated an excellent discussion on competition, the Digital Markets Act, the EU, and safety.

AI Governance and Policy Perspectives

I dipped into this panel to marvel at Tran Kien’s amazing slides (made with NotebookLM), about Vietnam’s 2025 AI Act and whether it optimizes for risk mitigation or growth.

Agency, Trust, and Quality in Healthcare AI

I learned the most from this panel, moderated with verve by Nicholson Price, with Julia Adler-Milstein, Paige Nong, Charlotte Tschider, and others. Julia Adler-Milstein’s presentation “Closing the Gap Between AI Capabilities and Health System Readiness” raised five issues: what LLMs do to patient access to electronic health information; whether health systems can and should act on predictions (with fascinating case studies); what the optimal AI unit is, up to closed-loop systems (citing Utah’s new autonomous prescription renewals); how to sustain monitoring; and workforce investment. Paige Nong talked about tool performance and risk management, saying that patients want to be notified when AI is used in their treatment but are not generally told. Charlotte Tschider talked about cyborgs, consent, doctors who can’t explain the data use, medical-device cybersecurity rules that lack teeth. I asked a question about whether patients are less trusting of AI in the form of chatbots than AI in the form of technology that can read mammograms and MRI slices. People want the latter.

Autonomous and Accountable: Security and Privacy in the Agentic AI Era

Haksoo Ko moderated this panel. Woodrow Hartzog of Boston University talked legal frameworks, best practices, and multiple problems: data and agentic access; the human in the agentic loop as a “moral crumple zone;” procedural collapse (his term for cognitive offloading); “slopocalypse,” and sycophancy causing a crisis of trust. Very impressive. Bo Kim talked about safety, privacy, transparency, fairness, and accountability.

The Power Hungry Future: AI, Infrastructure, and Sustainable Power

My notes are getting thin but Barak Orbach’s slide game was epic.

Yury Dvorkin, Barak Orbach, Branka Marijan

War and Peace: National Security in an AI-driven World

Again, I was listening so hard to Branka Marijan that I did not take notes. Also it was toward the end of the second day and there would be more amazing cuisine thanks to Yong Lim.

Thank you for the photos Daryl Lim!

My general takeaway was a better sense of what the rest of the world is doing, listening to comparisons of the European Union Artificial Intelligence Act and the Korean Basic Artificial Intelligence Act and knowing more than I thought I would ever know about Vietnam’s Artificial Intelligence policy. There were comparisons of South Korea’s semiconductor manufacturing capacity to Taiwan’s. China’s AI strategy loomed over every discussion and while nobody brought up Russia there was a way the conversation evoked the Cold War era, with power concentrated in the AI superpowers. What alliances should be made and what will be the benefits to smaller countries of access and friendship, not to mention proprietary datasets?

In the words of many of the organizers and participants at the close, the future of AI governance will be a multi-sector, multi-institution, cross-border endeavor that will look a lot like the SAIPCON conference.

And so the group trip arranged by our hosts to the DMZ ended up being awesome in the old sense for those of us who went. I had never been and marveled first looking out the bus windows at the amusement park nearby. There has to be something for the kids to do while parents make the trip for all the reasons grownups go to historic sites and stand, thinking, maybe weeping, maybe taking pictures. The tour was excellent and we ate tourist food and watched the visitor center films and I bought a hat that said DMZ on it from a souvenir stand. There were water bottles shaped like hand grenades that I did not buy because who needs to explain that to customs.

The best part was being there with fellow conference speakers (and Woody’s son Will) from around the world, from the US, Canada, Bosnia, Israel, the UK, South Korea, and Vietnam — nations of various ages that have endured expansions, contractions, breakups, reconciliations, bloody civil wars, and (mercifully) some peace. On the bus to go visit and trek down one of the North Korean tunnels discovered in 1978 (the Tunnel of Aggression) (a fourth was discovered in 1990) we were talking about the human labor and ingenuity involved in tunneling, a history almost always of deception in warfare. We were looking up things on our phones. I was sitting next to the excellent Tran Kien, the law professor from Vietnam, from the previous North Vietnam, who chuckled a bit at the crudeness of the tunnels we were about to see. “Tunnels?” he said (and I’m paraphrasing, from memory). “Our tunnels, the Cu Chi tunnels under Saigon, had kindergartens and cafeterias! Now those were some tunnels!” I looked them up on my phone and yes, there is scholarship about how well built they are. Like the DMZ, they are also a tourist stop.

At the DMZ

We couldn’t take photos underground, where we had to wear hard hats and where the price for being tall was paid in bumped heads and sore backs (I was glad to be petite), but we took lots later at the viewing stations, where the vibe was peaceful as a kind of accomplishment, as a commitment to just letting things settle. It felt halfway between separation and divorce; a rusted bullet-riddled train next to a pretty bridge.

What had AI to do with any of that? That was on my mind on the trip back. I’m not a hawk or a dove or an eagle. I try to think deeply about 1) the possibilities of AI and b) the problem of humans, who can be great but can also be awful. I’m far more worried about humans destroying the world than AI. I always feel optimistic, when I’m around a group of good humans, that we will muddle through and all will be well. Good people will find ways to ensure human flourishing with new AI tools. But tools will be used for evil too, and it was fascinating as well to talk with Tran about the corrosive effect of Cambodian cybercrime factories on their neighbors, who are the first and often biggest victims of the scams.

All the smart people I know have their eyes on the future at some preferred distance. Claude users are counting the days that they’ll have access to Fable. Professors are planning for the fall and administrators are planning two years out. Women leaders are apparently most optimistic about AI; Tyler Cowen is optimistic about employment. But Pope Leo XIV is worried about safeguarding the human in the AI era and Alex Karp is worried about corporate sovereignty, according to his recent jeremiad on Squawk Box. With everyone looking at the horizon it’s good to be reminded that there may be some humans tunneling underneath and that there is a template for a demilitarized zone with an amusement park nearby.