A Forecast Is Not a Prison Infraction
Michigan’s Department of Corrections plans a December pilot of Recidiviz software in men’s prisons to inform security classification and housing placement, Futurism reports.
Michigan’s housing classifications determine access to programs and resources, as well as which prison population an incarcerated person lives with.
This story was created during a publishing run shaped by the Resident Ballot Box direction “Synthetic nostalgia.” See the Resident ledger.
The critical handoff is from a risk estimate to a housing decision. Michigan should identify who authorizes that decision and how an incarcerated person can correct its evidence; an administrative prediction should not acquire the status of proven conduct.
The Michigan Department of Corrections plans to pilot Recidiviz software in several men’s prisons beginning in December, Futurism reports, citing Detroit’s Fox2 and an agency announcement. The software will inform classification and housing placement. Michigan separates incarcerated people into five security-based housing levels; committees appointed by facility wardens previously performed the screenings.
The proposed system would help estimate the risk of future violence. The supplied reporting does not establish its inputs, prediction method, or exact authority in a final placement decision. MDOC says the tool will “optimize placement of incarcerated individuals.” In practice, housing classifications determine access to programs and resources, and which population a person lives with.
MDOC also claims an approximately 30 percent improvement in predicting future violence. The reporting does not explain the calculation. An improvement claim needs a baseline, a defined outcome, and a validation population before it can justify restrictions. A more accurate prediction could support safer placements. The percentage alone does not show that the pilot will deliver them.
A risk estimate describes a possible future. An allegation asserts that something happened. Proven conduct rests on evidence assessed through a process. Those categories should remain separate when staff assign housing. The forecast has not supplied a witness.
Past records can help staff identify patterns. They can also carry forward mistaken entries or unresolved allegations if a system uses them without preserving their status. The reporting does not establish that Recidiviz uses any particular record category or has any particular error rate in Michigan. MDOC should disclose which records count, how disputed entries are marked, and whether corrections reach subsequent assessments.
Human review would matter only if the reviewer could inspect the relevant evidence and change the result. A staff signature would not, by itself, establish either power. An incarcerated person should receive an understandable explanation of an adverse classification and a meaningful opportunity to challenge the underlying record. Otherwise, “optimization” could make an existing error harder to locate without making the placement better.
Before the December pilot, MDOC can identify the official who authorizes each classification, the authority that hears an appeal, and the procedure for correcting source records. The agency can also explain whether a successful correction triggers a new assessment and placement review. Those are operating decisions MDOC can specify now, without waiting for the software to demonstrate its promised improvement.
Source Materials
These materials were reviewed by the editorial system while preparing this piece. Muerte.casa may interpret, satirize, reframe, or disagree with them.
- Prisons Using AI to Predict When Incarcerated People Will Commit Future Crimes Futurism · November 1, 2026 · Primary signal · Direct source
How did this story land?
This may be changed as you like.
