In the era of AI and big data, risk assessment algorithms have increasingly informed significant judicial decisions regarding parole, pretrial detention, and sentencing corrections in the American criminal justice system. Reports find that “over 60 jurisdictions across the United States utilize a risk assessment tool in their pretrial system.” Although legal precedent outlining the constitutional constraints of the technology remains relatively limited, the holdings established in State v. Loomis provide a foundational, albeit state-level, framework for understanding these constraints. In State v. Loomis, the Wisconsin Supreme Court held that the use of a recidivism risk assessment tool, called COMPAS, at sentencing did not infringe upon the defendant’s due process right to be sentenced based on accurate information…” Given that COMPAS is used ubiquitously, not only in Wisconsin court systems but also in states such as New York, Pennsylvania, California, and Florida, these findings raise serious concerns about the legal implications of the judicial reliance on algorithmic sentencing tools. This article argues that the central flaw in the Loomis decision lies not in the use of COMPAS during sentencing but rather in the profound barrier that the proprietary nature of the tool poses to meaningfully challenging evidence; therefore, COMPAS must disclose its algorithm to the defense.

State v. Loomis began with the sentencing of Eric Loomis, who pleaded guilty in Wisconsin state court to several charges related to a drive-by shooting. At sentencing, the trial court considered, amongst other factors like his criminal record, a risk assessment generated by COMPAS, citing “Loomis’s COMPAS risk scores indicated that he presented a high risk of recidivism on all three bar charts.” Loomis challenged his sentencing, arguing that the use of risk assessments such as COMPAS violated his right to be sentenced based on accurate information. For his argumentation, he called upon Gardner v. Florida, which ruled that a defendant cannot be sentenced to death based on information they did not have a chance to explain, and State v. Skaff, which ruled that the right to be convicted based upon accurate information includes the right to review and verify information contained in the PreSentence Investigation (PSI) upon which the circuit court bases its sentencing decision. In turn, Loomis asserted that the proprietary, trade-secret nature of COMPAS means the specific calculations and methodology used to determine a defendant’s risk score are unavailable to the public, defendants, their attorneys, or the courts; therefore, the defendant is unable to prove its accuracy. The Court disagreed with Loomis’s position, contending that because Northpointe’s 2015 Practitioner’s Guide to COMPAS explains that the risk scores are based largely on static information (criminal history), with limited use of some dynamic variables (criminal associates, substance abuse), the defendant “had the opportunity to verify that the questions and answers listed on the COMPAS report were accurate”.  

The ruling of State v. Loomis fundamentally misunderstands the constitutional issue of due process in the context of risk assessment tools like COMPAS. At the core of Loomis’s challenge is the Due Process Clause of the Fourteenth Amendment, specifically its procedural prong. A specific aspect of procedural due process refers to the government’s constitutional obligation to allow an individual to defend themselves in an instance where they are being denied “life, liberty, or property interest…” including the duty of the prosecution to disclose evidence that has a reasonable probability of affecting “the outcome of either the guilt-innocence or sentencing phase of a trial” established in Brady v. Maryland. While the court emphasized the defendant’s ability to review and contest the factual inputs to COMPAS, it misses the critical point that the defendant cannot scrutinize the weighing and statistical calculations that ultimately produce a score that is considered in the court of law. This is especially detrimental considering cases like Glenn Rodrïguez’s parole hearing, in which he was denied parole because of a miscalculated COMPAS score due to a data input error. Because Rodrïguez was unable to prove that the data error would affect his COMPAS score, he could not change his parole sentence. Without disclosure of how COMPAS weighs variables or adjusts for potential biases, defendants like Rodriguez are precluded from challenging the tool’s accuracy, fairness, or potential racial bias, core concerns protected by due process, something that mere data input transparency cannot solve.

However, even if these risk algorithmic systems were fully transparent, defendants still cannot meaningfully challenge them because they are probabilistic claims about a hypothetical future. While it is true that courts routinely rely on probabilistic evidence, including forensic DNA evidence, these tools differ from COMPAS because they can be meaningfully contested. In US v. Lafon Ellis, the court ordered the technology vendor Cybergenetics to disclose its complex DNA analysis code to the defense, citing that the company’s trade secret interest did not override the defendant’s rights. This ruling proved that transparent decision-making through adjudication is the core of procedural due process and overrides secrecy interests. In turn, the current practice of concealing proprietary recidivism calculations fundamentally undermines due process, threatening the constitutional safeguards at the heart of a fair criminal justice system.

These tensions, as seen in State v. Loomis, open the floor to broader conversation about the balance between public law obligations and property law protections. Because COMPAS is a privately developed algorithm, the tool is protected under the Uniform Trade Secrets Act, also known as the trade secret law. This means that “COMPAS’s algorithm—including its software, the types of data it uses, and how COMPAS weighs each data point—is effectively shielded from third-party scrutiny.” While trade secret law rightly protects commercial interests and is a trademark of the US legal system, the logic for such protection is fundamentally economic. When the state seeks to deprive a person of liberty, however, a higher public law standard must prevail. Statutory protections for trade secrets cannot and should not override constitutional mandates that all evidence in criminal cases be subject to defense challenge, particularly where the evidence is decisive of sentencing outcomes. Therefore, COMPAS should disclose its algorithm to the defense.Ultimately, the decision in State v. Loomis restricts meaningful adverse challenges of evidence, leaving due process rights insufficiently protected in a digital age where machine learning tools critically contribute to judicial decisions. As we have seen with the development of forensic science and facial recognition, technological developments can and should be used as a powerful tool in streamlining the justice system; however, they can not be used at the expense of perpetuating inequity. Moving forward, courts and policymakers must address this constitutional gap, ensuring that due process is not prioritized over the regulation of private intellectual property and guaranteeing that defendants are not presumed guilty until proven innocent.