Risk assessment instrument (RAI)

A risk assessment instrument is a statistical or algorithmic tool used in criminal justice to estimate an individual's likelihood of a future outcome, most often reoffending, failing to appear in court, or violating supervision conditions, and to translate that estimate into a score or risk category. Courts, parole boards, and pretrial services use these scores as one input, alongside professional judgment, into bail, sentencing, or release decisions.

Instruments are typically built by fitting a statistical model, historically logistic regression, sometimes newer machine-learning methods, to historical case data and validating it against actual recidivism rate outcomes. This differs from unstructured clinical judgment, the prior standard: it yields a reproducible score, though the training data reflects patterns from the justice system that generated it, including any disparities in policing and charging.

Independent studies and journalism have found some widely used instruments produce higher false-positive rates for Black defendants than white defendants at the same score, discussed as algorithmic bias; some developers dispute the framing, noting that reasonable fairness definitions cannot all be satisfied by one model at once. Accountability concerns also center on limited Explainable AI transparency where a tool is proprietary, and on scores read as more objective than they are. Responsible AI guidance generally holds that a score should inform, not replace, human judgment, with periodic revalidation against local outcomes.