Rikers Island prisoner risk classification system increases violence 50 percent

Occurred: December 2019

A USD 27.5 million risk-based data analytics system designed to reduce violence at Rikers Island prison, New York, contributed to assaults and other attacks there increasing by almost 50 percent, according to a ProPublica investigation. 

Created by consulting company McKinsey, the system used an algorithm based on factors including age, possible gang affiliation, and any prior history in jail to determine where to house people behind bars with the least risk for confrontation.  

In April 2017, McKinsey reported that violence had dropped over 50 percent in the 'Restart' housing units at the prison. But documents and correspondence analysed by ProPublica discovered that the numbers were 'bogus' and the relevant units stacked with inmates thought to be compliant and unlikely to get into fights or to attack staff.

It also transpired that that IntelWatch, a data analytics system designed to predict gang violence and future violence at the prison, was 'riddled with bugs and errors', did not work, and was never used, despite having been paid for. Furthermore, McKinsey consultants and corrections officials used encrypted messaging app Wickr to communicate and share documents in a deliberate  effort to avoid transparency and accountability. 

McKinsey argued the programme positively impacted the areas of the jail in which it was piloted, and that its consultants were not involved in or aware of efforts to 'improperly skew or reduce the reported levels of violence in the Restarts.'

Operator: New York City Department of Correction
Developer: McKinsey

Country: USA

Sector: Govt - justice

Purpose: Reduce violence

Technology: Classification algorithm
Issue: Accuracy/reliability; Effectiveness/value

Transparency: Governance; Marketing

Investigations, assessments, audits 🧐

Page info
Type: Incident
Published: September 2023
Last updated: November 2023