Code & Cops: The Lucrative World of Police AI

The realm of police funding, from initial data collection to critical decision-making processes, represents a significant financial frontier.

I found myself outside a prominent glass and brick edifice in Fort Worth, Texas, where thousands had convened for an event heralded as “the future of policing in the digital age.” As a member of the press, my entry was restricted. However, through conversations with attendees at nearby locations, I gained insight into the technologies being showcased and discovered that artificial intelligence (AI) is poised to fundamentally reshape American policing.

The central promise of AI at this year’s International Association of Chiefs of Police (IACP) Technology Conference revolved around automating routine, yet legally significant, police tasks. This mirrors a common business pitch: delegate mundane “busywork” to machines, freeing humans for more impactful responsibilities. Yet, in law enforcement, automating seemingly minor administrative duties—such as meticulously completing police reports or reviewing a suspect’s history—carries profound implications for individuals' lives.

The conference showroom in May featured a range of AI products, including facial-recognition cameras, automated license plate readers, body cameras, chatbots for non-emergency 911 calls, gunshot detection systems, drones, and tools for report writing. This embrace of automation by the industry persists even as the public increasingly questions the diminishing human police presence in communities.

Police departments are progressively entrusting their decision-making processes to algorithms. Numerous tech startups are now marketing AI to law enforcement as a sophisticated, automated command center—a central digital intelligence capable of processing vast amounts of data, often gathered by other surveillance and automation tools from the same vendors, to optimize resource allocation. Intriguingly, this technological push is not met with universal enthusiasm even among police officers.

“A lot of it is sales gimmicks that don’t actually deliver on what the promise is,” noted Abrem Ayana, a police captain in Brookhaven, Georgia. Due to the nascent nature of this technology and the absence of comprehensive federal oversight or standardized industry regulations, officials like Ayana frequently have little recourse but to accept companies’ assurances regarding the safety and efficacy of their products.

For decades, police departments have utilized technology to analyze data and, theoretically, inform field decisions. However, some prominent past initiatives notably failed. Programs such as CompStat and PredPol (short for “computer comparison statistics” and “predictive policing,” respectively) were early attempts to mitigate human error through purportedly unbiased statistical analysis. Instead, these experiments often exacerbated the very issues they aimed to resolve. Crucially, while these early efforts did not usher in a new era of unbiased policing as proponents hoped, human decision-makers still retained ultimate control.

The current marketing narrative for this new generation of AI products suggests that past failures stemmed from a lack of objective, real-time data. AI, in theory, can bridge this gap by significantly increasing the volume of public safety data collected and the depth of its analysis. However, many public safety advocacy groups and legal experts caution that introducing “black box algorithms” into law enforcement will erode transparency and accountability, particularly at a time when public trust in police is already severely strained.

Jason Truppi, a former FBI special agent specializing in cybercrime, articulated that police forces are overwhelmed by an immense volume of data. Truppi, who wore Meta Ray-Ban Smart Glasses and spoke with energetic corporate terminology, cofounded ForceMetrics in late 2020. The software company offers an “AI-powered decision-assist platform, enabling public safety agencies to increase operational efficiency and better serve their communities in real time,” as detailed on its LinkedIn profile.

According to Truppi, the record-keeping systems used by police departments over the last two decades—encompassing everything from emergency call logs to parole records and body camera footage databases—have collectively led to an unmanageable information overload. He stated, “All the systems of record [used by police departments] are essentially antiquated.”

“We don’t use the ‘p word’ at all, because it failed.”

ForceMetrics provides police departments with a platform called Velocity, which its website claims “uses AI to turn overwhelming amounts of public safety data into clear, actionable insights.” In the lexicon of police-tech, Velocity functions as a Real-Time Crime Center (RTCC). First implemented by the New York City Police Department over two decades ago, RTCCs are designed to consolidate police data from various sources—such as 911 dispatch, CCTV cameras, and license-plate scanners—to furnish officers with a summary of what to anticipate upon arriving at a scene. The underlying premise is that greater access to real-time data reduces the likelihood of officers acting on “guts and guns,” a euphemism Truppi uses for situations that escalate negatively and result in fatalities.

Historically, RTCCs were managed by human analysts responsible for collecting, organizing, and relaying incoming digital data to officers on patrol. However, as Truppi indicates, the proliferation of new data-collection technologies in policing over the years has rendered it virtually impossible for any department to effectively manage this deluge of information. By 2019, the NYPD alone was collecting approximately two years’ worth of body camera footage weekly, according to a transcript from a 2019 Committee on Public Safety hearing—a volume far too great for even the most diligent human to meaningfully analyze.

Modern RTCCs, such as Velocity, are engineered to rapidly identify patterns within vast datasets, aiming to enhance situational awareness for officers. Truppi contends that the “unfortunate events” that have severely eroded public trust in police departments in recent years, particularly during the pandemic, can largely be attributed to a deficiency in what he terms “a data-driven approach” to policing.

Nina Loshkajian, a fellow at the New York University Center on Race, Inequality, and the Law, expresses skepticism regarding this assertion. She stated, “The reality is that police departments had already been using predictive algorithms, which companies touted as data-driven, for years before calls to defund the police revved up in 2020. These algorithmic systems did not prevent violent encounters between police and civilians then, and we shouldn’t be tricked into thinking they’ll make a meaningful difference in the future.”

Truppi’s company faces competition from two dominant entities in the contemporary police-technology industrial complex: Motorola Solutions and Axon Enterprise. Both firms develop their own RTCCs and produce many of the data-collection and surveillance technologies on which these centers rely.

In early 2024, Axon—originally known as TASER—acquired the surveillance technology company Fusus to launch its own RTCC, officially branded as Axon Fusus. By this point, Axon was already a prominent provider of stun guns, body-worn cameras, and automated license plate readers. The company’s portfolio also includes Draft One, a popular AI-powered report-writing tool; drones for police departments via its Axon Air program; and its proprietary AI chatbot.

Axon and Motorola are among a select group of companies vying to effectively monopolize the entire modern police technology ecosystem, from data acquisition at crime scenes to the strategic decision-making capabilities of AI-powered RTCCs. Police departments frequently enter into multiyear contracts with these providers, who often offer free trial periods for new technology and utilize “sole-source procurement agreements,” enabling them to continue selling new products without facing competitive bids from other vendors.

Andrew Guthrie Ferguson, a professor at Georgetown University Law School and author on policing and technology, observes, “We’re seeing a gold rush into selling [AI] technology to police with the promise that it will all make their jobs easier and more efficient.”

This “gold rush” has also drawn considerable interest from external investors. According to tech entrepreneur Amber Schroader, whom I interviewed during the event in Fort Worth, approximately a quarter of the attendees on the conference showroom floor were from “equity firms looking to invest in the latest tech.” She noted, “That was a surprise.”

The sales pitch has clearly resonated.

For instance, Draft One and similar AI-powered report-writing tools hold significant appeal, especially considering that the average police officer dedicates 40 percent of a typical shift to report writing, as per a 2024 study by Axon. Many of these reports pertain to routine incidents like traffic stops and noise complaints. John Mackey, a patrol sergeant with Colorado’s Avon Police Department, which uses Field Notes—an AI report-writing tool by Truleo—remarked, “We didn’t sign up to sit behind a keyboard. That wasn’t why I became a police officer.”

Draft One incorporates design elements aimed at ensuring human oversight. For example, the system intentionally leaves certain details blank, requiring officers to manually complete them. The platform is built on a modified version of ChatGPT, specifically trained for police report generation, and Axon claims it is free of hallucinations. Noah Spitzer-Williams, senior principal product manager at Axon’s generative AI division, has stated, “The creativity is turned down to zero.” However, this assertion should be viewed with considerable skepticism, as even leading AI research labs like OpenAI, Anthropic, and Google have yet to fully eradicate hallucination from their most advanced models. A notable incident earlier this year highlighted this challenge when Draft One reportedly described a Utah officer transforming into a frog, having apparently picked up audio from the Disney movie *The Princess and the Frog* playing in the background.

While such an incident might seem amusing, the real-world consequences of AI-written police reports could be gravely serious. When a human officer authors a report, they can be cross-examined in court regarding their state of mind, their rationale for including specific details, or their reasons for omissions. By its very nature, it is impossible to subject "black box algorithms" to the same level of scrutiny.

Axon and Motorola are part of a very small group of companies competing to effectively monopolize the entire modern police technology stack, from the collection of data at crime scenes to the strategic decision-making capabilities of AI-powered RTCCs.

In the initial iteration of Draft One, it was also impossible to ascertain which portions of a submitted report were AI-generated and which were human-authored, beyond an officer’s personal recollection. This was presented as a deliberate design choice, not a flaw. In a recorded roundtable discussion published online shortly after Draft One’s 2024 launch, Spitzer-Williams explained that the platform “by design” does not save an original copy of a report post-submission. He elaborated, “because [the] last thing we want to do is create more disclosure headaches for our customers and our attorney’s offices… it’s actually never stored in the cloud at all so you don’t have to worry about extra copies, you know, floating around.” Consequently, if a Draft One-generated report appeared in court and contained inaccuracies, attorneys or judges had no definitive way to determine whether the errors originated from the officer or the AI.

Axon spokesperson Victoria Keough confirmed that Draft One was updated in December to allow police departments “to retain and access the original, unedited AI-generated narrative.” This modification was implemented “as [law enforcement] agencies, prosecutors, policymakers, and legislatures have established clearer expectations and requirements for AI-assisted report writing.”

Brandon Garrett, a professor at Duke University School of Law who has researched the implications of AI systems for due process, views this technology with apprehension. He finds the concept of “making up data—which is what generative models do—to be used in court, is really, really troubling.” Garrett argues, “We would never tell a police officer, ‘Just be creative and come up with a story about what you saw at the crime scene.’ Of course not: They’re supposed to objectively record as best as they can and document what they saw at the crime scene. But generative models are designed to create.”

Following the 2008 financial crisis, LA police chief Charlie Beck, inspired by personalized shopping algorithms used by companies like Walmart and Amazon, advocated for police departments to adopt similar tools for crime prediction. Starting in the 2010s, “predictive policing” programs became widespread across American cities. However, these algorithms, far from ushering in an era of fairness and justice, often produced the opposite effect. Because the models were trained on historical crime data, they perpetuated existing biases embedded within that training data, under the guise of objective analysis.