Michigan prepares prison-scoring pilot as developer rejects the AI label

Michigan’s upcoming prison-classification pilot is meant to improve housing decisions—not hand them to AI, its developer says. A November 10 Fox News article highlighted that distinction and an unresolved performance claim: Michigan expects roughly 30% better predictions of prison violence, but its announcement supplied neither the underlying results nor an explanation of that estimate.

A December trial with daily consequences

The Michigan Department of Corrections announced the redesigned system on October 28. Testing is scheduled to begin December 1, 2026, at selected men’s prisons. The department says the pilot will establish how well the tool performs before it considers wider use. Recidiviz, a nonprofit technology organization, helped develop it.

Classification determines much more than a label in a file. It affects where someone lives, how freely they can move, how closely staff supervise them and which activities they can access. Michigan oversees about 32,000 incarcerated people and uses four general-population security levels: I, II, IV and V.

Higher levels impose tighter movement restrictions. People at lower levels may have more vocational-training and higher-education opportunities, though Michigan says Violence Prevention Programming remains available at every general-population level. A revised score can therefore affect both security and the opportunities available during a sentence.

The department also wants reviews to keep pace with changing records. Its current inputs are collected on a schedule; the redesigned process will support ongoing reviews incorporating updated information. That change is intended to let staff reassess a person’s classification without relying solely on the next scheduled collection.

The AI description meets a rules-based scoresheet

Early coverage described something different. Futurism’s November 1 article called the planned deployment a major new AI system. On November 2, Recidiviz published an explanation rejecting that characterization. It said both Michigan’s existing classification instrument and its replacement use fixed factors and points—not AI.

The developer says the redesign began with an analysis of 10 years of Michigan classification data. Corrections experts reviewed potential scoring factors. Those with statistically significant relationships to violence inside prisons were retained. Other factors were individually considered for removal or as possible grounds for staff to override a score.

According to Recidiviz, the resulting formula changes which factors count and their weight. It also gives behavior over time a role in the path toward less restrictive custody:

• Recent disciplinary incidents carry more predictive weight than older ones, allowing sustained good conduct to help reduce a person’s security level.
• Completing certain programs can support a lower-security classification because participation is associated with less violence inside prisons.
• Changes in a person’s record can trigger an automatic review flag, reducing the need for staff to track those changes manually.

The software pulls information from department records and adds the assigned points. Staff can inspect each factor and its value, then make the final placement decision. They can override the score for circumstances the data may not capture, including medical needs or personal safety concerns.

The result is a scoresheet with a set number of points for each factor.

Recidiviz, November 2 explanation of Michigan’s classification system

A prediction still needs somewhere to lead

That inspectable formula does not settle the performance question. Michigan’s approximately 30% improvement remains an expectation, not a published pilot result. The department’s announcement does not identify the measure behind that percentage or show the underlying comparison. The December trial is explicitly intended to test performance before broader deployment.

Housing capacity presents another constraint. Byron Osborn, president of the union representing Michigan corrections officers, has warned of a shortage of higher-security beds, according to Fox News. His objection concerns whether the department has appropriate places for people already classified as higher risk—not simply whether a new score can identify them.

Michigan has announced 870 additional high-security beds under its prison-safety initiative. But more than 400 are planned in new units at Oaks and St. Louis correctional facilities, with construction expected to take about two and a half years. Those projects operate on a much longer timetable than the classification pilot scheduled to begin in December.