NYPD Stop and Frisk Litigation

In August 2013, after a nine-week trial, the U.S. District Court for the Southern District of New York issued a liability opinion finding that the NYPD's stop-and-frisk practices violated the constitutional rights of the plaintiff class. The Court found that the NYPD violated the Fourth Amendment, which requires Terry stops to be based on reasonable suspicion, and also found New York City liable for a pattern and practice of racial profiling during Terry stops in violation of the Equal Protection Clause of the Fourteenth Amendment. 

Alongside the liability opinion, the Court issued a separate remedies opinion requiring immediate reforms to NYPD's training, documentation, supervision, and monitoring, and appointed an Independent Monitor to oversee implementation and ensure constitutional compliance.

In February 2021, the Court approved the research studies designed to leverage body-worn camera footage. This study is one of those two.

The Study

The study uses AI tools and techniques, including machine learning and natural language processing, to computationally analyze BWC recordings and identify key indicators of constitutional compliance in what NYPD officers say and how they say it during encounters with civilians. 

It was approved alongside the CUNY Institute for State and Local Governance study, with each designed to contribute distinct but complementary analyses of the NYPD's compliance with Fourth and Fourteenth Amendment principles. The ISLG Report, filed in May 2025, used retired New York State judges to measure constitutional compliance, providing reliable legal determinations. The Stanford study, by contrast, uses machine learning techniques to analyze BWC recordings, focusing on what officers say and how they say it during encounters.

Body-worn cameras, required by the Court's remedial orders, now provide an objective record of what officers actually say during encounters — information largely invisible in stop reports and other administrative records but essential for assessing constitutional compliance.

The study focuses on two areas where AI and language-based methods provide the clearest insights for compliance monitoring. First, it evaluates the NYPD's compliance with New York's four-level framework for classifying police interactions established by People v. De Bour, with a specific emphasis on Level 3 stops and detentions that require reasonable suspicion, mirroring the Fourth Amendment's requirements. 

Second, it analyzes the NYPD's consent search practices for compliance with the Fourth Amendment, which requires that consent be voluntary, an inquiry that turns in part on the language officers use. To assess compliance with the Fourteenth Amendment, the study analyzes differences in officer language during encounters with civilians of different races and ethnicities.

The Data

The study draws on NYPD encounters from three primary sources. For understanding De Bour documentation and compliance, we analyze a sample of 1,702 encounters between March 16 and May 15, 2022, drawn from data that are the focus of the ISLG Report, and 1,156 encounters assessed by the Monitor team between 2022 and 2024. For our consent search analysis, we examine the full set of consent requests documented on stop reports in 2023 that we were able to match to footage, totaling 1,770 encounters and 3,695 videos.

Together these samples cover a range of encounter types, time periods, and sources of expert review. The ISLG Sample includes encounters independently assessed by retired New York State judges for constitutional compliance. The Monitor-Assessed Sample draws on three years of the Monitor team's quarterly audits of Level 2 investigative encounters. The Consent Search Sample covers the full population of consent search requests officers documented on stop reports throughout 2023.

The NYPD records millions of investigative encounter videos each year, but quarterly Monitor audits and ComplianceStat meetings review only a few hundred BWC videos. 

Computational methods make it possible to analyze large volumes of this footage in ways that would be impossible through manual review. For each sample, BWC recordings are matched to administrative records, transcribed, and preprocessed into a structured format suitable for analysis. Automatic speech recognition tools convert BWC audio into text transcripts of officer and civilian speech, and a subset of recordings were transcribed manually for model training, evaluation, and comparison.