November 2, 2026

Michigan Begins Testing New Prisoner Classification System

Michigan will commence testing a new system for classifying prisoners on December 1, aiming to enhance predictions of future violence by approximately 30%. This effort follows a notable increase in assaults on U.S. prison staff, rising by 77%, and assaults among incarcerated individuals, increasing by 54% from 2019 to 2024, as reported by the Michigan Department of Corrections (MDOC).

The pilot program is set to launch in specific men’s facilities. The classification system affects prisoner movement, supervision, and access to activities and programs. The nonprofit technology organization, Recidiviz, which played a role in developing the system, confirmed that this tool does not implement artificial intelligence (AI) for decisions related to security levels or prisoner release.

Michigan oversees roughly 32,000 incarcerated individuals and has embarked on revamping its classification method due to the surge in nationwide prison violence. Recidiviz was brought on board last year to assist in creating the new system, with the objective of potential wider application pending trial outcomes.

Functionality of the New Classification System

Historically, Michigan employs four security categories: Levels I, II, IV, and V; Level III is absent. These classifications influence a prisoner’s freedom of movement, supervision degree, and activity involvement. Higher-security levels entail stricter controls, whereas lower levels grant broader access to opportunities like vocational training and higher education.

The updated system will integrate real-time data for continuous analysis, moving away from periodic data collection alone. MDOC anticipates a 30% increase in predictive accuracy regarding potential violence, however, the public statement lacks clarity on how this figure was derived.

Clarification from Recidiviz on AI Usage

Recidiviz does offer AI-supported solutions through its correctional platform, but they emphasized that AI is not used for determining security classifications in Michigan. A spokesperson from Recidiviz clarified to Newsweek that AI is not a component of the tools used for influencing security levels or prisoner release decisions.

Michigan’s Objective for Safety

MDOC Director Heidi E. Washington emphasized the importance of aligning security levels with the current behaviors and needs of prisoners to safeguard both prisoners and staff. The pilot, utilizing evidence-based historical data, is anticipated to strengthen facility security and provide a dynamic, objective tool.

MDOC plans a gradual implementation, starting with this pilot to assess the system’s effectiveness before considering broader application.

Concerns Regarding Algorithmic Risk Assessment

Using algorithms in the criminal justice arena has previously prompted debates concerning transparency and bias, as seen in a ProPublica inquiry into a distinct classification system employed in Louisiana. This system, which evaluated reoffending and parole eligibility, varied from Michigan’s security classification approach.

Consequently, the state’s announcement leaves public inquiries on various aspects unanswered such as the data utilized for assessments, methods determining accuracy improvements, potential staff overrides, and bias detection mechanisms.

The resolution of these issues remains crucial as Michigan contemplates making this pilot a permanent fixture in prisoner classification processes.

TAGS: