Prison walls block the web. Yet inmates peck away at typewriters, scribbling prompts for ChatGPT. They hand notes to visitors. Free-world contacts feed answers back. The New York Times reports Nick Browning, locked up at Jessup Correctional Institution in Maryland, types queries on an old manual machine. He wants AI to sharpen his legal push for better cancer screenings behind bars. No internet. No direct access. Just ingenuity.
Browning isn’t alone. Hype seeps into cellblocks. Prisoners crave the tool everyone outside buzzes about. They draft motions. Hunt case law summaries. Even plot post-release job hunts. But rules clamp down hard. Most facilities ban online tools outright. Fears of contraband schemes or escape plots run high.
So. Workarounds multiply.
Inmates dictate prompts aloud during rare visits. Loved ones relay them to public AI interfaces. Responses smuggled in on paper. Or memorized. One prisoner in California described feeding prompts through a lawyer’s paralegal. Output returned via mail. Slow. Risky. Effective.
From Reentry Drills to Surveillance Scares
Not all AI touches stay underground. Programs test official channels. Northeastern University’s Prison Project in Massachusetts equips inmates with generative AI chatbots on secure tablets. No web needed. Justice Trends Magazine details how participants build résumés. They practice job interviews. Draft business plans. ‘Learning how to use ChatGPT and incorporate it into my business plan was huge,’ one said. Staff bottlenecks vanish. Reentry skills flow 24/7.
D’Amore-McKim School at Northeastern drives this. Students guide sessions at Suffolk County House of Correction. Tools like ChatGPT turn lived experience into ATS-friendly bullet points. Confidence builds. One inmate sketched a brand logo via AI prompts. Release looms brighter.
But benefits collide with risks. Prisons already deploy AI against inmates. Securus Technologies trains models on years of calls. MIT Technology Review explains pilots scan real-time chatter for crime flags. Texts. Emails. Video visits. Kevin Elder, Securus president, says tools launched in 2023. Inmates pay for monitored lines. Privacy erodes.
LeoTech partners similarly. Facilities record, transcribe, flag calls. Berkeley Technology Law Journal flags attorney-client pitfalls. AI can’t spot privilege alone. Hallucinations hit 1% in top tools like OpenAI’s Whisper. With 2 million incarcerated, that’s 20,000 botched transcripts daily. ‘38% of hallucinations include explicit harms,’ a study warns. Trials turn unfair.
And courts grab chatbot logs. No privilege shield. In U.S. v. Heppner, a New York judge ruled defendant chats discoverable. Orrick AI Law Center analyzes: neither attorney-client nor work product applies. Prosecutors peek freely.
Missouri cases pile up. Ryan Schaefer confessed vandalism to ChatGPT minutes after. Logs doomed him. Another suspect detailed fires, child abuse plots to bots. Evidence handed over.
Regulators circle. States target companion chatbots. California mandates self-harm protocols, minor safeguards. Troutman Privacy notes SB 243 effective January 2026. New York requires crisis referrals. Michigan bars minors from unregulated bots.
Future of Privacy Forum tracks 98 bills across 34 states. FPF. Patchwork looms. Oregon, Washington enact soon.
NIJ briefs tout chatbots for supervision. State Justice Institute. Reduce admin loads. Free staff for high-value work. But demand human oversight. Privacy. Safety checks essential.
Ethical shadows linger. AI offers therapy-like talks. Lacks compassion. AI & Society cautions: platforms personalize content, not care. Inmates vulnerable. Manipulation risks rise.
Prisoners push on. Browning types relentlessly. AI bolsters his screening fight. Northeastern grads reentry plans. Securus scans deepen control.
Balance tips slowly. Tools empower. Traps tighten. Facilities weigh access against threats. Lawmakers draft rules. Inmates adapt.
One thing clear. AI breaches bars. Changes incarceration from within.
WebProNews is an iEntry Publication