AI Parole and Probation Risk Assessment Discrimination 2026
Long after a bail hearing or sentencing decision is over, a second generation of algorithmic scores takes over — setting how closely a parolee or probationer is supervised, whether a parole board grants early release, and how a violation gets handled. These post-conviction tools carry their own disparity findings and their own due-process gaps, distinct from the pretrial risk-scoring debate most compliance guides focus on.
A Different Decision Point Than Bail or Sentencing
Pretrial risk tools get most of the public attention because they intersect with the presumption of innocence. But actuarial risk scoring shows up again after conviction, in instruments like the Ohio Risk Assessment System (ORAS) and the Level of Service Inventory-Revised (LSI-R), which corrections departments and parole boards use to classify supervision intensity, recommend caseload assignments, and inform release and revocation decisions for people already sentenced.
The stakes are different from a bail hearing, but not smaller. A high-risk classification can mean more frequent check-ins, stricter travel and employment conditions, and a lower threshold for a technical violation to escalate into a revocation hearing that sends someone back to incarceration — a consequence with its own body of research showing that supervision violations, not new crimes, drive a large share of prison admissions nationally.
What the Disparity Research Actually Found
Validation studies of widely used supervision instruments have found that criminal-history-heavy inputs — number of prior arrests, prior supervision violations, age at first arrest — produce higher average risk scores for Black and Hispanic supervisees even when controlling for the underlying offense. Because arrest and violation records themselves reflect decades of uneven policing and enforcement intensity across neighborhoods, a tool trained on that history can reproduce the same disparity it's used to manage going forward.
Unlike pretrial tools, which have drawn national attention through a small number of widely cited investigations, supervision-level instruments are adopted state by state and department by department, with validation quality varying significantly. Some departments have commissioned local disparity audits; many continue using a vendor's original national validation study without confirming it holds for their own supervised population.
Where the Legal and Policy Pressure Is Building
Revocation Hearing Due Process
HIGH RISKWhether a supervisee can meaningfully contest a risk score that shaped a hearing officer's revocation recommendation, absent a formal right to inspect the tool's scoring logic
Supervision-Level Disparate Impact
MEDIUM RISKClaims that facially neutral inputs produce racially disparate supervision-intensity classifications, echoing employment-discrimination disparate-impact theory
Parole Board Reliance Without Disclosure
MEDIUM RISKBoards using a risk score as a factor in early-release decisions without explaining its limitations or error rate to the person being scored
Vendor Validation Claims
MEDIUM RISKWhether a tool validated on one state's supervised population was adequately re-validated before another department adopted it
What Supervision Agencies Should Document
The direction of travel for departments continuing to use these tools mirrors what's emerged in pretrial risk scoring, adapted to the supervision context:
- A validation study specific to the department's own supervised population, not just the vendor's original national sample
- A documented policy on how much weight a risk score carries in supervision-level decisions and revocation recommendations, versus an officer's individualized case assessment
- A standing disclosure to supervisees explaining that a risk score influenced their classification and its known limitations
- Outcome tracking by demographic group across supervision level, violation rate, and revocation rate to catch disparity the original validation study didn't surface
For vendors selling into corrections and community-supervision markets, the emerging expectation is disclosure of input categories and validation methodology to the purchasing department, plus support for that department's own local disparity audit rather than treating the original validation study as permanently sufficient.
Compliance Checklist for Corrections Departments and Vendors
Relevant to parole boards, probation departments, and risk-assessment tool vendors.
Audit your AI product's compliance exposure
RatedWithAI helps tech teams understand their compliance posture across accessibility, privacy, and AI regulation requirements. Start with a free scan.
Scan Your Product for Free →Frequently Asked Questions
Are ORAS and LSI-R the same tool used for pretrial bail decisions?
No. ORAS and LSI-R are actuarial instruments built for post-conviction supervision classification and case planning, distinct from pretrial tools like COMPAS or the Public Safety Assessment, which score defendants before trial for bail and release decisions. Some jurisdictions use different tools for each stage; others reuse elements of the same scoring logic across both, which is itself a point of scrutiny.
Does a probation department need a defendant's consent to use a risk-scoring tool?
No — these tools are generally adopted as part of a department's standard supervision procedure, not something an individual can opt out of, which is precisely why validation quality and disclosure requirements are the primary compliance lever rather than consent.
Can a supervisee get a copy of their own risk assessment score and the factors behind it?
Policies vary significantly by state and department. Some jurisdictions provide the score and general factor categories to the supervisee or their counsel on request; others treat the full scoring output as an internal case-management record. Where a score is expected to influence a revocation hearing or parole decision, due-process arguments for disclosure are strongest.