Michigan Prisons Prepare to Pilot an AI Tool That Predicts Inmates' Risk of Future Violence and Feeds Security Classifications
JUST IN: Michigan prisons preparing to deploy AI that predicts incarcerated people's risk of committing future violent acts, using the results to determine their security classifications.
Context
It is one of the clearest cases of a US state letting AI risk scores shape how people are treated in custody, which raises hard questions about bias and accountability.
Risk scoring tools have been used in US criminal justice for years and have long drawn criticism over bias. Polymarket reported on Nov 1 that Michigan prisons are preparing to deploy an AI system that predicts the risk of future violent acts and uses the results in security classifications. The details below come from an X news summary of posts and were not checked against an official Michigan Department of Corrections release.
Deep dive
Polymarket (@Polymarket) posted on Nov 1 that Michigan prisons are preparing to deploy AI that predicts incarcerated people's risk of committing future violent acts, using the results to determine their security classifications. The post passed 100,000 views, with about 1,700 likes and more than 200 replies at scan time.
According to an X news summary of the discussion, the Michigan Department of Corrections plans to start a pilot on December 1 in some men's facilities, working with the nonprofit Recidiviz. The summary says the system would update risk scores based on crime details and behavior history, that officials expect violence predictions to improve by about 30 percent, and that the plan follows a 77 percent rise in assaults on staff and a 54 percent rise in fights between prisoners since 2019. It also says humans keep the final say on classification and housing. These figures come from that summary and have not been confirmed against an official source for this story.
Reactions split. Some people welcomed a more objective way to place prisoners, while critics compared it to the idea of punishing crimes before they happen and raised concerns about bias in the data the model learns from.
What to expect next
Watch for an official statement or documents from the Michigan Department of Corrections confirming the start date, the facilities and how the model is evaluated. Civil liberties groups are likely to ask for audits of accuracy and bias across racial groups. If the pilot goes ahead in December, other states may look at the results, and legal challenges over due process are possible if scores affect housing or privileges.