AI Behind the Walls: Insights from 100,000 Inmate Conversations with Generative AI (PID041)

12pm – 12.30pm WEST, 23 April 2026 ‐ 30 mins

Parallel Workshops

What happens when you place cutting-edge generative AI directly into the hands of incarcerated individuals? Over seven months and more than 100,000 real conversations, we discovered an unexpected mix of hope, vulnerability, creativity, and risk. This session shares the results of two groundbreaking inmate-facing AI pilots at a 640-bed jail - a Companion Chatbot designed to reduce loneliness and a Career Guide Chatbot built to support reentry planning. Together, they generated remarkable engagement: an average of 1.4 sessions per user per day, with conversations lasting about eight minutes.

But the story is more than numbers. We’ll explore what incarcerated users actually talk about with AI, the emotional and behavioral patterns that emerged, and what we learned from the best and worst interactions - including the top 10 most common conversation themes and the moments that tested the limits of safety and moderation. Attendees will also gain a first look at a proposed framework of foundational principles for safe, ethical, and human-centered inmate-facing AI, including guardrails that prevent access to prohibited topics such as weapon-making, poison-making, alcohol production, and escape planning.

This presentation invites correctional leaders to imagine a future in which AI can support rehabilitation, mental wellbeing, and reentry readiness without compromising safety, security, or dignity. It’s an honest, data-driven look at one of the most provocative questions in modern corrections: Can AI be a force for good behind the walls - and what does it take to get it right?