Michigan’s Department of Corrections plans to introduce an AI-supported system designed to estimate which prisoners could pose a potential risk of violence. The assessment would be used to place inmates in one of five security categories, according to reporting by Futurism.
The department expects the system to produce risk assessments that are 30 percent more accurate than current methods. A pilot project is scheduled to begin in September 2026 at several men’s prisons. The stated aim is to reduce a recent increase in attacks on employees and violence between incarcerated people.
The software was developed by the corrections department in partnership with Recidiviz, a nonprofit organization that originated at Google. Department director Heidi E. Washington said the pilot would use scientific and historical data to create a dynamic and objective tool intended to improve safety in the facilities.
However, the department has not disclosed which data the AI system will process. That information is important because the results could affect prisoners’ conditions during their sentences. Depending on their classification, inmates may receive different programs or resources and could also be assigned to different cellmates.
Errors or uncorrected biases in the system could therefore lead to stricter conditions and potentially poorer preparation for life after release. A ProPublica investigation into a similar risk-assessment program in Louisiana found that the data used for classification included racial and socioeconomic biases, with negative consequences for the corrections system.
Sources: t3n.