The Hidden Dangers of Bail Reform

How Pretrial Risk Assessments Decrease the Effectiveness of Cashless Bail and Pretrial Release Systems

In the push for bail reform, pretrial risk assessments have emerged as a supposed solution to eliminate surety/cash bail systems and promote fairer pretrial release decisions. These algorithmic tools, designed to evaluate a defendant’s risk of failing to appear in court or reoffending, aim to shift away from traditional bail practices.

By analyzing factors like criminal history, age, employment, and socioeconomic data, they generate risk scores to guide judges on whether to detain or release individuals without requiring monetary bail. Advocates claim these assessments offer an objective path to cashless bail, reducing jail populations and addressing biases in the criminal justice system. However, growing evidence shows that pretrial risk assessments are unreliable, discriminatory, and counterproductive, often perpetuating racial inequalities and threatening civil liberties. 

Biased Foundations: How Flawed Data Sabotages Bail Reform and Cashless Bail Initiatives

The core issue with pretrial risk assessments lies in the biased data they rely on, which directly impacts pretrial release outcomes and undermines bail reform goals. Historical criminal justice records are often criticized as being tainted by disproportionate policing in communities of color. This has the potential to inflate arrest rates for Black and Latinx individuals compared to their white counterparts for similar offenses. This distorted data trains algorithms to label marginalized groups as higher risk, even when their behavior doesn’t justify it.

Cashless bail risk assessments are just barely better than a coin flip

A pivotal 2016 ProPublica investigation into the COMPAS tool—a common pretrial risk assessment used in bail reform efforts—highlighted these flaws. Analyzing over 7,000 defendants in Broward County, Florida, the study revealed only 61% accuracy in predicting recidivism, scarcely better than chance. More critically, it uncovered racial disparities: Black defendants were almost twice as likely to be wrongly flagged as high-risk (44.9% false positives vs. 23.5% for whites), while white defendants were more often misclassified as low risk despite reoffending (47.7% vs. 28% for Blacks). Even after adjusting for priors and age, Black defendants faced a 77% higher chance of a high-risk score for violent crimes. This stems from indirect racial proxies like zip codes or family history, which embed discrimination into pretrial release decisions and contradict the equity promised by cashless bail and bail reform.

Leading data scientists speak out against “black box” algorithms

These problems extend beyond COMPAS. In 2019, 27 leading researchers from MIT, Harvard, Princeton, NYU, UC Berkeley, and Columbia issued a statement on the “Technical Flaws of Pretrial Risk Assessments,” warning that these tools jeopardize bail reform. Key concerns include:

• Inaccurate Risk Definitions: Vague metrics for “risk” (e.g., flight or danger) misalign with real outcomes, leading to unfair denials of pretrial release.
• Failure to Predict Serious Crimes: Tools can’t reliably identify risks for violence, conflating minor and major offenses, which hinders effective pretrial systems.
• Biased Inputs: Arrests and convictions mirror enforcement biases, skewing results against people of color and perpetuating cycles that bail reform seeks to break.
• Oversimplified Categories: Labels like “high-risk” ignore uncertainties, causing judges to over-deny pretrial release and rely less on individualized bail assessments.

Experts argue that no amount of adjustments can erase these flaws, as they’re rooted in inherently biased data. Consequently, pretrial risk assessments amplify detention rates for African-American and Latinx defendants, thwarting the unrealistic overpromise of bail reform. 

Wider Risks: How Pretrial Risk Assessments Fuel Mass Incarceration and Erode Bail Rights

Pretrial risk assessments not only discriminate but also pose systemic threats to the criminal justice system. They encourage detention based on probabilistic scores rather than concrete evidence, potentially increasing jail populations instead of reducing them. This new form of algorithm-based incarceration creates new injustices in pretrial release processes. For instance, a high-risk label can result in pretrial detention without proper due process, violating the presumption of innocence which is central to a fair bail system.

Additionally, these tools enable “digital prisons” through surveillance like electronic monitoring for those granted pretrial release, extending government oversight into daily life. The lack of transparency in proprietary algorithms, such as COMPAS, further prevents accountability, allowing biases to persist and undermine trust in bail reform efforts. 

Growing Resistance: Civil Rights Advocates Challenge Pretrial Risk Assessments in Bail Reform

Opposition to pretrial risk assessments is mounting among civil rights leaders. In July 2018, over 100 organizations—including the ACLU, NAACP Legal Defense Fund, Color Of Change, and The Leadership Conference on Civil and Human Rights—released a joint statement condemning these tools. They asserted that assessments “reproduce harm” by institutionalizing racial biases in pretrial release decisions, inflating risks for Black and Brown communities based on flawed policing data. This leads to higher detention rates.

This widespread criticism signals a consensus: Pretrial risk assessments don’t neutralize inequalities—they intensify them. 

Pathways to Genuine Bail Reform: Moving Beyond Algorithms for Equitable Pretrial Release and Cashless Bail

Given the discrimination, inaccuracies, and rights erosions tied to pretrial risk assessments, it’s time to rethink their place in the justice system.

As evidenced by ProPublica, top researchers, and civil rights coalitions, pretrial risk assessments are a hazard, not a help. If we want a pretrial system that is truly fair and balanced, we must reject black box algorithms and automated biases and allow judges to have the discretion they need to do the job they were trained to do.