01
Factual background & dispute
- The presentence investigation report included results from a COMPAS risk and needs assessment.
- The defendant challenged the proprietary model's lack of transparency and argued that its use of gender data violated due process.
- The sentencing court also considered independent factors, including the severity of the offense, prior criminal history, and supervision records.
02
Core issues & judicial focus
- Whether the inability to review the proprietary algorithm violates due process rights
- Whether risk scores may be used in individualized sentencing
- What limitations must be imposed when risk assessment models consider factors such as gender
03
Judicial finding & holding
- The court affirmed the post-conviction order denying resentencing.
- COMPAS could be considered with warnings about its limits and independent sentencing factors. Scores could not determine incarceration or sentence length, or decisively determine whether community supervision would be safe and effective.
- The court did not establish that the tool was unbiased or permit algorithm-only sentencing. COMPAS is a recidivism risk tool and is not described here as generative AI.
04
Practical risk implications
01High-impact algorithmic scores must be accompanied by usage limitations, known error rates, target population parameters, and update histories.
02Final decision-makers should document the actual role of algorithmic scores in their determinations.
03Vendor trade secret claims increase the burden on authorities to provide explanations, facilitate cross-examination, and safeguard procedural rights.