Announcing Our 2026 Scholarship Winner: Marcus Castillo
By the SEUT Board | July 2026
This year's SEUT Scholarship drew more than 900 applications from high school students across the country, easily surpassing the 600+ applications from our first contest in 2025. Students shared thoughts on what ethical technology means in their lives, and what they are working on to promote ethical technology in their schools and communities. Our board read and scored all of them, built a shortlist of finalists, and spent weeks in interviews before arriving at a winner.
That winner is Marcus Castillo, a rising senior at the Salpointe Catholic High School in Tucson, Arizona.
Marcus's essay stood out for its clear-eyed reflection on balance and responsibility in technology. Rather than writing in the abstract about AI and ethics, he wrote about a real tension he had run into in his own work: the tools he built to help people were capable of harming the very people he set out to help, and he had to decide what to do about it.
Who Marcus Is
Marcus founded AID4AZ, a youth-led coalition working to expand civil legal aid in Arizona. Through the Smart Border Initiative, he has built AI tools meant to make legal help more accessible: document generators, bilingual resource maps, and visa guidance bots. Alongside that work, he has researched how AI trained on past court decisions can absorb and repeat the same biases as the system it learns from, work he pursued with faculty at Princeton's Center for Information Technology Policy. This summer, he's joining Harvard Law School's Access to Justice Lab as a student researcher.
Congratulations are also due to this year's twelve finalists, who impressed our board with their own work and perspectives across privacy, mental health, governance, and community advocacy: Deborah Joshi Samuel, Rilyn Rodgers, Xavier Perry, Evan Sidio, Navya Arora, Allizon Carrera, Alanya Jamison, Long Nguyen, Felix Lee, Gabrielle Gelber, Naval Shah, and Advait Kothuri.
In His Own Words
Marcus's winning essay, published in full below.
I didn't set out to work in technology ethics. I set out to help people who couldn't afford lawyers, then realized that the tools I was building could hurt the same people I was trying to help.
When I founded AID4AZ, a youth-led coalition expanding civil legal aid in Arizona, the problem felt straightforward: most people facing legal crises can't afford representation, and the gap between needing help and getting it is often money. So when I began building AI tools through the Smart Border Initiative — document generators, bilingual resource maps, visa guidance bots — it felt like progress. Faster, cheaper, more scalable than anything a nonprofit could deploy alone.
I started asking what happens when AI learns from the legal system.
Working with faculty at Princeton's Center for Information Technology Policy, I analyzed over 5,000 judicial opinions to find patterns in how courts rule. What I found were patterns in who loses. A model trained on past court decisions doesn't correct for bias. It learns from it. Consider a tenant facing eviction who can't afford representation. She loses. That outcome enters the data. The model trains on it. The next tenant, in similar circumstances, is predicted to lose too, not because the law says so, but because the system learned to expect it. Automation, in that case, isn't neutral. It's efficient in the wrong direction.
That didn't make me want to stop building. It made me want to build differently and understand the problem rigorously before deploying at scale.
This summer I'm joining Harvard Law School's Access to Justice Lab as a researcher. The work sits at the intersection of two questions: how AI tools can help self-represented litigants navigate courts, and how to expand legal aid without reproducing the inequities embedded in legal data. The answer to both is the same: treat fairness as a technical requirement from the start, not a value you bolt on after launch. My goal coming out of that research is concrete: a framework for auditing legal AI tools before they reach vulnerable populations, and eventually a model legal aid organizations can adopt.
Long term, I want to work where all three levers meet: building tools, shaping the policy that governs them, and practicing law in a way that keeps both accountable. No single person fixes a broken system, but the people working on access to justice in AI need to understand what it feels like to sit across from someone who got a denial letter they can't parse and have nowhere to turn. I've sat in those rooms. You can't theorize your way out of that.
The gap between who the legal system serves and who it was built for is large. AI will either widen it or help close it, and that depends entirely on the choices made before a model ever touches data. Every assumption baked into a training set is a decision about whose losses get normalized. I know whose losses I'm building against. And I know what it costs when no one does.
What's Next
Alongside this year's scholarship, we're launching the SEUT Chapter Founders Council, inviting student leaders like Marcus and this year's finalists to start SEUT chapters at their own schools. If Marcus's essay resonated with you, or you're a high student who wants to build something similar in your own community, we'd love to hear from you.