AI analysis of NYPD body-camera footage finds underreported stops and racial disparities

Researchers at the University of Michigan, UC Davis, and Stanford analysed over 4,600 NYPD body-worn camera encounters using machine learning and natural language processing. The study found evidence of stops that were not formally documented, racial disparities in officer interactions, and widespread use of unclear language during consent searches. AI models classified stops with over 80% accuracy and identified undocumented stops at over 70% accuracy, allowing review teams to uncover more than half of problematic encounters by examining just 25% of footage.

In this domain: AI in digital forensics: 68% of DFIR professionals now use AI in investigative workflows · AI Act deadline extension: compliance programmes pausing now risk exposure if Omnibus is not formally adopted · 70+ civil organisations call on Meta to halt FRT smart glasses as ACLU demands disclosure of law enforcement discussions