1. State v. Loomis (Wisconsin Supreme Court, 2016)

State v. Loomis

Background

This is the most important global case on algorithmic sentencing tools. Eric Loomis was sentenced in Wisconsin, and the judge used a risk score generated by COMPAS (Correctional Offender Management Profiling for Alternative Sanctions). COMPAS is a proprietary algorithm that predicts the likelihood of reoffending.

Legal Issue

Whether using a secret, proprietary algorithm in sentencing violates:

• Due process
• Right to challenge evidence
• Transparency in judicial decision-making

Court’s Decision

The Wisconsin Supreme Court allowed COMPAS to be used but imposed strict limits:

Key findings:

• COMPAS cannot be the sole basis for sentencing.
• Judges must treat it as one factor among many.
• Defendants must be warned about:
  • The tool’s limitations
  • Its proprietary nature (cannot fully challenge it)

Significance

The court accepted algorithmic input but emphasized:

• Transparency problem (black-box risk scores)
• Bias risk (racial and socioeconomic bias concerns)
• Judicial discretion must remain primary

2. R (Bridges) v Chief Constable of South Wales Police (Court of Appeal, UK, 2020)

R (Bridges) v Chief Constable of South Wales Police

Background

Although not a sentencing case, this case is crucial for understanding algorithmic governance in criminal justice.

South Wales Police used Automated Facial Recognition (AFR) technology in public spaces to identify suspects.

Legal Issues

The court examined whether the algorithmic system violated:

• Privacy rights
• Equality laws (risk of biased targeting)
• Public law requirements (clear rules for discretion)

Court’s Findings

The Court of Appeal ruled the system was unlawful at the time because:

1. Lack of clear legal framework

Police had too much unstructured discretion in deploying the algorithm.

2. Equality concerns

The system lacked adequate safeguards against bias affecting protected groups.

3. Data protection issues

Insufficient explanation of how individuals were selected or matched.

Significance for sentencing algorithms

Even though this was not sentencing, it established principles relevant to sentencing tools:

• Algorithmic tools must be transparent and legally structured
• Public authorities must ensure accountability for automated decision systems
• Discretion cannot be hidden inside technology

3. State of the Netherlands v SyRI Case (District Court of The Hague, 2020)

State of the Netherlands v SyRI

Background

The Dutch government used a system called SyRI (System Risk Indication) to detect welfare fraud. It combined large datasets and produced “risk scores” identifying individuals likely to commit fraud.

Legal Issues

Whether the algorithmic system violated:

• Right to privacy (Article 8 ECHR principles)
• Right to fair process
• Transparency in government algorithm use

Court’s Decision

The court struck down the system entirely.

Key reasoning:

• The algorithm was too opaque for citizens to understand or challenge
• It created a disproportionate intrusion into private life
• Risk scoring lacked meaningful safeguards against discrimination

Significance

This case is often cited as one of the strongest judicial rejections of black-box predictive governance systems, reinforcing that:

• “High-risk scoring” systems must be explainable
• Bulk data profiling requires strict proportionality
• Citizens must be able to challenge algorithmic outputs

4. State v. Malenchik (Indiana Supreme Court, 2010)

State v. Malenchik

Background

Indiana courts used risk assessment tools during sentencing to evaluate rehabilitation potential and recidivism risk.

Legal Issue

Whether sentencing courts may rely on structured risk assessments when determining punishment.

Court’s Decision

The Indiana Supreme Court allowed risk tools but with caution:

Key principles:

• Risk assessments are permissible advisory tools
• They must not replace judicial reasoning
• Sentencing must remain individualized

Significance

This case is important because it predates modern AI systems but establishes foundational doctrine:

• Algorithmic scoring can inform sentencing
• But it cannot override individualized justice
• Judges must explain how they weigh algorithmic input

5. People v. Wesley (New York, DNA/forensic algorithm case line of reasoning, 2000s)

People v. Wesley

Background

While not a sentencing algorithm case directly, this case is widely used in legal reasoning about scientific and computational tools in criminal trials, including probabilistic and algorithmic evidence.

The case involved the admissibility of DNA statistical analysis used to influence sentencing and conviction severity.

Legal Issue

Whether probabilistic algorithmic evidence can be trusted and admitted in court.

Court’s Approach

The court accepted the evidence but emphasized:

• Scientific methods must be reliable and testable
• The defense must be able to challenge methodology
• Courts must understand error rates

Significance for sentencing algorithms

This case laid groundwork for modern algorithmic governance principles:

• Even probabilistic tools affecting sentencing must be explainable
• Courts must understand algorithmic uncertainty
• Expert scrutiny is required before reliance

Core Legal Principles Emerging Across These Cases

Across all five cases, courts consistently return to a few controlling ideas:

1. Transparency Requirement

Algorithms cannot remain fully “black box” if they influence sentencing outcomes.

2. Right to Challenge

Defendants must be able to contest:

• Inputs
• Methodology
• Outputs (risk scores)

3. Judicial Discretion Cannot Be Delegated

Algorithms may assist but cannot replace human sentencing judgment.

4. Bias and Equality Concerns

Courts are increasingly sensitive to:

• Racial bias in training data
• Socioeconomic bias in risk scoring systems

5. Proportionality and Privacy

Especially in European cases, mass data profiling must be strictly necessary and proportionate.

Overall Conclusion

Sentencing guideline algorithms sit in a legally sensitive space:

• Courts are generally not rejecting them outright
• But they impose strict conditions:
  • Explainability
  • Limited reliance
  • Procedural fairness
  • Judicial oversight

The direction of law across jurisdictions is clear:
algorithms may assist sentencing, but they cannot govern it.