Michigan turns to AI tools as prison violence rises

The Michigan Department of Corrections plans to pilot a partially automated inmate classification system developed with Recidiviz, a Utah-based nonprofit technology organization. Starting Dec. 1 at selected prisons, the system would replace manual worksheets with assessments updated in real time. It would consider crimes, expert evaluations and behavior in custody to help determine appropriate security levels for the state’s 32,000 inmates.

Department Director Heidi Washington said the tool is expected to improve predictions of future violence by approximately 30%. A department spokesperson said AI would not decide an individual’s security level or release. Staff would review calculated assessment scores and make final classification decisions. The module forms part of a broader Recidiviz platform that also supports restrictive housing reviews and prison transfers.

Corrections officers’ union president Byron Osborn questioned whether better assessments would address the underlying shortage of high-security beds. Officers also link rising violence to chronic understaffing, lockdowns and inmates being placed below their assessed security level. Recent suspected cellmate killings have intensified those concerns. The department has announced plans to add 870 high-security beds, but Osborn said most would not be ready for years and immediate action was needed.