What are AI risk-assessment tools used in criminal sentencing?

A tool with roots older than modern generative AI

Risk-assessment tools in criminal justice actually predate the current wave of generative AI, having developed out of earlier actuarial and statistical models used to predict outcomes like recidivism or failure to appear in court. These tools take in data about a defendant — such as criminal history, age, and other case-specific factors — and generate a score or risk category meant to estimate the likelihood of a particular outcome, such as reoffending if released. Courts in a number of states have incorporated these scores into decisions about bail, sentencing recommendations, and parole.

How they’re generally meant to be used

The stated purpose of these tools is to inform a judge’s decision-making by providing a standardized, data-driven data point, not to replace human judgment with an automated verdict. In practice, this means a risk score is typically presented alongside other case-specific information — the nature of the offense, the defendant’s individual circumstances, victim input, and more — for a judge or parole board to weigh together, rather than functioning as a binding, automatic outcome.

Significant controversy and legal scrutiny

Despite this intended role, risk-assessment tools have drawn substantial criticism and legal challenges. Much of the concern centers on evidence that some tools have produced racially disparate outcomes, and on the fact that many of these tools rely on proprietary scoring methodologies that defendants and their attorneys cannot always fully inspect or challenge — raising due process concerns. This has led some jurisdictions to reconsider, limit, or add additional oversight to how these tools are used, while others continue to rely on them as one input among several.

Bottom line

AI-driven risk-assessment tools generate scores meant to help inform, not replace, judicial decisions about bail, sentencing, and parole, but their use remains legally and ethically contested due to concerns about bias and the transparency of how scores are calculated.