James O’Donnell reports in MIT Technology Review on AI scanning prisoners’ communications

James O’Donnell reports in MIT Technology Review on AI scanning prisoners’ communications

Published December 1, 2025, James O’Donnell’s MIT Technology Review newsletter reports that Securus Technologies is piloting an AI model to scan prisoners’ calls, texts, and emails in the hope of predicting and preventing crimes. Securus president Kevin Elder told the publication that the company trained models on recorded calls, including calls from the Texas prison system, and is working on models for other states and counties. The newsletter does not report whether the pilot has predicted or prevented crimes. Prisoner rights advocates say the system enables invasive surveillance, while courts have specified few limits to that power. The concern is not only what the model can detect, but how broadly prison communications may be monitored and what limits govern that monitoring. The newsletter leaves unclear what safeguards apply to the pilot and how its predictions are used.

Publication: ba9be736-2167-2e1f-eef2-568d7892245b

attribution — uncertain: The reading is one newsletter issue, so it counts as one piece. James O’Donnell is credited only on the prison-calls item, not as the newsletter's author. Calling it 'James O’Donnell’s MIT Technology Review newsletter' overstates his role. Smallest repair: 'an item by James O’Donnell in MIT Technology Review’s The Download newsletter.' The title and claims are otherwise correctly attributed to Elder, the advocates and the publication.

Read the full story. —James O’Donnell

summary — pass: The summary accurately conveys the development: a company is piloting AI scanning of prisoners' calls, texts and emails. It also preserves the advocates' surveillance objection and the court-limits point. The reading omits that Securus is a US telecom company that trained models on its own call database, but the central meaning stays clear. The question is a genuine open issue that follows from the evidence.

A US telecom company trained an AI model on years of inmates’ phone and video calls

date — pass: The stated date of December 1, 2025 matches the UTC day of originalPublished (2025-12-01T13:10Z). The Deep Dive links to other Download issues may be later additions, but the reading does not use them. There is no hindsight framing. The pilot is not presented as having produced outcomes. This is a publisher page, so no archive disclosure is needed.

is now piloting that model to scan their calls, texts, and emails in the hope of predicting and preventing crimes.

support — pass: The claims match the source: the pilot is aimed at predicting and preventing crimes, Elder says models were trained on recorded calls including Texas, other states and counties are being worked on, and advocates object while courts have set few limits. The sentence 'The concern is not only what the model can detect...' is a mild unlabeled interpretation, but it stays within the advocates' stated surveillance concern. The noted gaps about safeguards and outcomes are accurate.

It created one model, for example, using seven years of calls made by inmates in the Texas prison system, but it has been working on models for other states and counties.

media — pass: No media was selected, and no verified candidates were supplied.