Release Assessment recommends release alike across races in NYC

In 251,917 New York City arraignments from 2021 to 2023, CJA's updated Release Assessment recommended release on recognizance for 87% of Black, 88% of Hispanic and 85% of white people.

  • 2 for a person

    Most people assessed got a release recommendation, 88% from 2021 to 2023, at rates that barely differed by race, a measurable improvement in the advice that judges receive about them.

  • 2 for society

    Across the city the spread in recommendation rates between racial groups was cut in half on the development data, from 9.4 to 4.3 percentage points.

Depth runs from 1, a small gain, to 5, a life saved.

Status
Realized benefit.
Evidence

foundstudiedreportedclaimed

An independent research report supports the benefit.

Size

0.24 =2+4+1+520×3+110

Scale2The deepest rating above
Reach4251,917 arraignments got a Release Assessment recommendation from 2021 to 2023, and 88% of them were recommended for release. Source
Duration1Each person gets one recommendation at one arraignment, and the sources do not document how long release lasted. Source
Distribution5The update cut the spread in release recommendation rates across races in half, from 9.4 to 4.3 percentage points, on the development data. Source
Probability3Realized.
Frequency1A person is assessed once, before arraignment, and the recommendation goes to the judge at that single hearing. Source
Compared with

The 2003 CJA Release Assessment, on the same 2009 to 2015 arrests. Measured.

Without the system
9.4 percentage points spread in release recommendation rates across races
With it
4.3 percentage points
The system’s part
Not isolated. The sources compare the updated tool with the 2003 tool on historical data and report recommendation rates by race. No source compares release outcomes with and without the tool. The sources call the tool a statistical model with eight weighted factors and do not describe machine learning.
Learns from data

Shown. CJA says that two research teams built statistical models on 1.6 million arrests and assigned points to each factor by its strength in the final model.

“Next, the two teams worked together to build statistical models. The collaboration yielded a Release Assessment model with eight predictive factors, with points assigned to factors based on their strength in the final statistical model.” Source

Later
Persisted, 1 October 2025. DCJ's 2025 report shows that the tool was still in daily use through 2023, and that judges followed its release recommendations in only 30% of violent felony arraignments.

Download data Cite Suggest a correction

Type of benefit
Increased Inclusion
System
CJA Release Assessment (updated 2019)
Made by
New York City Criminal Justice Agency with Luminosity, Inc. and the University of Chicago Crime Lab
Used by
New York City Criminal Justice Agency, for the New York City Criminal Courts
Where
New York City, United States
Date
Kind
Aggregate result
Broad benefit
Liberty protection (criminal justice)

Sources

  1. Report, by authors with no tie to the system.

    Goh LS, Rempel M, Weill J (2025). Alignment with New York City's Pretrial Release Assessment: Results for the Five Boroughs. Data Collaborative for Justice at John Jay College of Criminal Justice, October 2025.

    The assessment recommended 87% of Black, 88% of Hispanic, and 85% of white people for ROR.

  2. Report, by the organization that used it.

    Peterson RR (2020). Brief No. 46: CJA's Updated Release Assessment. New York City Criminal Justice Agency, September 2020.

    The difference in the rates of recommendation for ROR across all race/ethnicities was cut in half (from 9.4 percentage points under the 2003 CJA Release Assessment to 4.3 percentage points in the updated version), as was the difference between sexes (from 7.1 percentage points to 3.4 percentage points).

  3. News report, by authors with no tie to the system.

    Pinto N (2025). NYC Judges Have a Tool to Help Them Rule on Pretrial Detention. Why Are They Ignoring It to Make Racially Skewed Decisions? Hell Gate, 16 October 2025.

    Significantly, judges are deviating from the recommendations of the tool more for Black and Hispanic people, meaning that even though judges have a racially neutral mechanism for considering pretrial detention, they're choosing to disproportionately lock up people of color anyway.

Details

AI system

Workflow

Two research teams fitted the tool to 1.6 million New York City arrests from 2009 to 2015 and kept the eight factors most tied to court appearance. CJA interviews people before arraignment and assessed 98.0% of those held for arraignment in the first months.

Follow-up

Conditions for replication

The factors must stay predictive as court practice changes, which is why CJA planned a revalidation. Judges must follow the recommendation for the even treatment to reach release decisions, which the 2025 report found that they did unevenly.

The entry

Sector

Criminal justice (pretrial release).

Purpose

The New York City Criminal Justice Agency built the updated Release Assessment to tell judges at arraignment how likely a person is to return to court, with less pretrial detention and smaller racial disparities.

Cite this benefit

Peter H. Vartanian. (2026). Release Assessment recommends release alike across races in NYC (AIBD-1222). Artificial Intelligence Benefits Database. https://petervartanian.xyz/aibd/records/AIBD-1222.html (archived at https://doi.org/10.7910/DVN/AADSII)

BibTeX Data What the fields mean