Michigan Prisons Deploy Recidiviz AI to Score Inmate Risk

Michigan's Department of Corrections will begin letting an AI tool from the nonprofit Recidiviz sort incarcerated people into five housing security levels, with a pilot launching in December at select men's facilities, Futurism reports.

The state's own claim for the tool is specific and unexplained: it will "improve MDOC's ability to more accurately predict incarcerated individuals' risk of future violence by approximately 30 percent." MDOC has not published the methodology behind that figure. The system replaces committee-based assessments previously conducted by facility wardens; Recidiviz itself began as a data organization project inside Google.

The precedent Futurism points to is Louisiana, where a ProPublica investigation into a similar risk-analysis system found the underlying data "contained inherent racial and socioeconomic bias." One documented case involved "a blind and wheelchair bound 70-year-old" classified as a moderate risk despite a clean disciplinary record, which triggered automatic loss of parole eligibility.

Housing designations determine program eligibility and the peer population an incarcerated person lives among for the duration of a sentence, so a scoring error compounds. Two researchers we track flagged the piece on day one.