How technology can make for fairer and more effective policing, prosecutions and jails
Over the last few years, many New York City justice reform efforts have effectively stalled. Recent reports from the federal monitor overseeing the City’s stop-and-frisk policies note continuing issues with police conduct during stops. New York City's call-diversion program saw a substantial portion of eligible calls fail to receive program services. And the City looks certain to miss the legally mandated deadline to close Rikers next year, with horrid conditions continuing to decimate the credibility of our justice system. In light of these delays, many advocates have grown impatient, calling on Mayor Zohran Mamdani to increase investment and redirect City resources to unblock these reforms.
Reform is difficult for many reasons, including inadequate staffing and tricky political and policy questions. But reforms also ask swamped public servants to do more. They impose information burdens: staff must scrutinize what happens more closely and carefully assess alternatives to the status quo, on top of their already demanding daily workloads. In theory, the City could hire legions of new public servants to handle this work, but doing so would be prohibitively expensive in a time of severe budget pressure.
Instead, recent advances in AI offer the City a cheaper and more effective way to tackle these information burdens, freeing public servants to focus on decisions that require their expertise and judgment. AI could help supervisors identify misconduct and exemplary behavior in vast archives of body-worn camera footage; help 911 call takers and prosecutors divert more cases away from traditional criminal resolution; and help prosecutors and public defenders organize enormous amounts of digital evidence that otherwise delay cases and prolong jail stints at Rikers. And so, at the same time the City tackles staffing issues and political logjams that block reform, it should also prioritize testing and deploying AI in these settings, giving public servants the tools they need to deliver on longstanding reform commitments.
Improving police conduct. One enduring concern for the federal monitor overseeing the NYPD's compliance with the 2013 Floyd ruling is whether officers are documenting all street stops. An obvious way to check whether stops are actually documented is to tap the NYPD's enormous body-worn camera footage archive. In theory, body-worn camera repositories provide a tremendous opportunity to measure officer behavior at scale. This vision is what drove support for their rollout across the country; indeed, one of the key conditions of the Floyd ruling was that the NYPD pilot the use of body-worn cameras.
But the NYPD lacks the personnel capacity to review most of this footage. As of 2019, about two years' worth of video was recorded and stored by the NYPD every week. Thorough review would require hundreds of new analysts to watch this footage full-time.
Instead, to identify unreported stops, the NYPD and the monitor typically ask staff to painstakingly review a small random sample of body-worn camera clips. Estimates of reporting rates across all recorded encounters depend heavily on what reviewers find in this small sample. For example, in the monitor's most recent report on unreported stops, reviewers found a single example of an unreported stop among 200 randomly sampled clips labeled as low-level encounters. That single example drove the monitor's estimate of nearly 2,000 unreported stops in low-level encounters in 2025. As the monitor notes in its report, if reviewers had happened to find one more or one fewer unreported stop in the sample — purely due to chance — the monitor's estimate of unreported stops citywide would have been a third higher or lower.
AI could help the NYPD more precisely estimate unreported stop rates and assess whether new policies improve compliance with the monitor's requirements. First, AI would review this footage at scale, rating clips by the likelihood they contain a stop. Next, reviewers would sample body-worn camera clips across these ratings, devoting more attention to clips with a high likelihood of containing a stop while still checking a few clips with a low likelihood. This statistically informed approach could substantially improve the precision of stop documentation estimates without requiring more staff review time. Indeed, researchers commissioned by the monitor released a report in April (which they wrote about in Vital City) demonstrating the promise of this approach.
Several private-sector vendors, including Truleo and Polis, have also built tools to review body-worn camera footage at scale. Randomized experiments in Aurora, Colorado, and Richland County, South Carolina, found that using these tools can improve some measures of officer communication, although these benefits have not been confirmed widely. And a study of two departments in Arizona suggests that officers can be receptive to this monitoring when it rewards good police work.
Recognizing this promise, the NYPD announced a pilot with Truleo in the fall of 2023. But the NYPD has not publicly shared results from the pilot or indicated whether it intends to proceed with a permanent contract. The City and NYPD should take advantage of these tools to better identify unreported stops, intervene when officers fail to comply with department policy, and recognize exemplary police behavior. These changes would help the NYPD and the City demonstrate the sustained improvements needed to end federal oversight.
Increasing diversion. New York City has struggled to divert nonviolent 911 calls, like those involving a mental health crisis, away from police response. Although the City's B-HEARD program responds to thousands of calls every year, these represent just a tiny fraction of all calls in the precincts the program covers. Generally, it's unclear how many calls can realistically be answered without police. Yet even if police are the ones responding, many people would benefit if officers had better options for referring them to effective treatment, shelter or other services.
The biggest barrier to call diversion is the limited availability of effective community-based care — a problem technology alone cannot solve. If the City addresses this barrier, the next hurdle will be helping 911 call takers quickly and accurately identify calls that are suitable for referral to outside services, and either send nonpolice responders or help officers connect people with these services. Researchers have begun using AI to support these referrals, both by training call takers with AI that simulates callers and by using AI to assess a call's risk in real time as it unfolds. Increasing the scale of this triage could help more people avoid the justice system.
Once someone has been arrested and their case has been referred for prosecution, the city's district attorneys can recommend people for diversion programs, which are intended to help them avoid incarceration and connect with services like drug or mental health treatment. But these programs are often underutilized, in part because it takes a lot of effort for busy prosecutors to assess a person's eligibility and make a referral. AI could quickly review case evidence to make preliminary eligibility assessments, nudge prosecutors to recommend diversion for those who appear eligible, and even complete low-risk referral paperwork to reduce the burden of the referral process. Making referrals easier — and reminding prosecutors when they are a good option — could help more people get the treatment they need and avoid a long stint on Rikers.
Reducing incarceration by speeding up case processing. For people held in pretrial custody in New York City, long case processing times mean more time on Rikers Island. Reports over the last 10 years have flagged that case processing delays inflate the Rikers population. The median felony case processing time increased over 30% between 2018 and 2022, while the city's average jail stay in 2022 was more than twice that of the five largest jail systems in the country. Fortunately, the City Comptroller estimated that reducing average processing times for non-murder felony cases by just four weeks could reduce the City's jail population by more than 1,000 people, making this a key target the City needs to hit before it is able to close Rikers.
One core cause of these delays is that prosecutors and defense attorneys have to wade through an enormous amount of digital evidence. As the world becomes increasingly digital, detectives compile a growing array of digital media on each case, including surveillance video, body-worn camera footage, social media logs, medical records, cellphone extractions and more. In a 2025 study, one attorney reported at least 10,000 pages of data per cellphone extraction, while another office received 620 hours of footage per attorney annually — nearly 30% of the entire workyear if attorneys were to watch this footage eight hours a day. In some jurisdictions, defenders now receive more than 40 times as much evidence every year as they did a decade earlier. (New York's recent discovery reforms also require prosecutors to disclose a more comprehensive range of evidence to defenders, further increasing the amount of digital evidence compiled for each case.)
While this explosion in digital evidence could theoretically give both prosecutors and defenders a better picture of what happened in each case, in practice it has caused repeated delays to case processing as they manually sort through everything that has been provided. Indeed, a 2019 pilot in Brooklyn found that delays were often caused by incomplete discovery materials or requests for extensions because attorneys did not have enough time to review all provided evidence.
Generative AI can speed up rote work while preserving attorney discretion for the numerous high-stakes decisions involved in prosecution and defense. For example, AI can automate compiling a list of all involved people on a case; classifying folders of unlabeled documents by their source and type; and placing all evidence on a common timeline. With these tools, attorneys could get up to speed on the evidence faster, allowing them to identify grounds for dismissal earlier, begin informed plea negotiations earlier, and be ready for trial sooner. Paired with better court scheduling, these tools could help criminal cases resolve faster and reduce the amount of time people spend languishing in jail while their cases proceed.
To be sure, using AI in these ways would present tricky challenges for the City. First, the City needs the capacity to implement and manage these interventions. While the City’s Public Interest Technology (PIT) Crews represent a promising step, the longer-term goal for the City should be to move beyond this consulting model by permanently embedding technically capable civilians across government, making modern technology a routine part of public-sector operations. Fortunately, massive recent layoffs have made it possible to recruit experienced technologists who are tired of supporting the tech sector's march toward “enshittification” — basically, the erosion of once-useful software in pursuit of higher profits — and who want to make a positive difference in the world instead.
As the City builds out its technical capacity, it can draw on a deep bench of public-interest talent to fill immediate gaps in expertise. Reasonable voices in the city's civic tech and advocacy community could assist in defining what responsible use looks like. Applied research labs across the city's numerous universities — including my own — can also provide pro bono technical assistance and credible independent evaluations of new interventions. Nonpartisan organizations like the Council on Criminal Justice have also established useful frameworks for responsible adoption. Developing in-house expertise and drawing on outside support would help the city rekindle its long tradition of responsible civic innovation in the age of AI.
A second challenge is the public's increasing wariness of the use of AI in justice settings. In February, an Amazon Super Bowl commercial reminded Americans that if the tech company can automatically track a lost dog in your neighborhood, it might be able to track you too. The public's uneasy reaction to this ad caused Amazon to void a collaboration with Flock Safety. This spiraled into a full-blown public relations crisis for Flock this summer, with multiple local governments canceling contracts with the company and people even removing Flock cameras in vigilante operations.
Here in New York City, inconsistent, spotty or opaque information about the NYPD's use of surveillance technology has only served to raise fears among the public that new technologies are being used in an irresponsible way. From 2007 to 2019, the City secretly procured a range of surveillance technologies — including cellphone-tracking tools — totaling nearly $3 billion in contracts. Although these contracts have started to come to light through public-records requests by the Legal Aid Society and the Surveillance Technology Oversight Project, the City continues, bafflingly, to fight disclosure of these contracts and their details. Although the City passed a law in 2022 requiring City agencies to disclose use of "algorithmic" tools on an annual basis, the mandate doesn't appear to include technology that most people would consider to qualify as AI, including its numerous license plate readers.
Beyond surveillance, generative AI presents other new modes of failure: hallucinations (incorrect or fake but convincing claims), sycophancy (AI saying whatever it thinks you want to hear) or deepfakes (synthetic images or video that are indistinguishable from authentic media). Automated systems also have the potential to perpetuate biases found in training data. Technology can also distract from sloppy police work and provide bad apples with a way to access information they shouldn’t see. There are plenty of promising ideas on how to reduce these risks. But we need to test these ideas in practice, iterating and learning which approaches reduce abuse and misuse.
In the meantime, the City will need to convince a skeptical and nervous public that it is taking credible steps to address these risks. The mayor has a talent for wonky, policy-dense communication; the City should consider adopting his approach to show — not tell — the public how its many diverse, talented and passionate civil servants are making responsible use of these new technologies.
Fears of AI-powered surveillance in many ways echo longstanding concerns about the overreach of the justice system, but the shiny cloak of new technology can obscure these historical parallels. As a society, we have always struggled to maintain the appropriate balance between the state's extreme power and safeguards against the irresponsible, overbearing or unequal use of this power — illustrated by the fact that four of the 10 constitutional amendments in the 1789 Bill of Rights guard against government overreach in criminal procedure.
But in the past, it's taken us a long time to strike an acceptable balance. Although the NYPD began adopting fingerprinting technology over a century ago, it took decades of litigation and mistakes before we devised mature policy on when fingerprints could be taken, how matches would be documented and challenged, and who could access this sensitive information. Nowadays, a clear majority of Americans consider fingerprinting to be acceptable police practice.
Today, AI is evolving so fast that we don't have decades to spare. We need a competent public sector that can draw from historical parallels — and learn from the ineffective or effective ways we've curbed overreach in the past — to address new threats posed by AI, all while finding opportunities for its responsible use. In New York City, an effort to use AI to kickstart stalled reform efforts would be a good place to start.
An earlier version of this piece appeared online in Vital City on Feb. 19, 2026. Alex Chohlas-Wood currently provides paid consulting services for the NYPD’s Racial Disparities Review Committee.






