I stood outside a massive glass-and-brick building in Fort Worth, Texas. Inside, thousands of cops gathered for what the brochures called “the future of policing.” I couldn’t go in. But the stories leaked out. And what came out was unsettling. AI isn’t just helping cops; it’s moving to seize the heart of American law enforcement.
The pitch at the International Association of Chiefs of Police Technology Conference was simple. Automate the busywork. Let machines handle the paperwork so officers can focus on “meaningful” work. But in policing, that “busywork” is the legal scaffolding that holds cases together. Automating police reports or case reviews doesn’t just save time. It changes the outcome of people’s lives.
The Black Box on the Beat
Look around the showroom floor. You’ll see facial-recognition cameras. Automated license plate readers. Drones. Gunshot detection. Chatbots for non-emergency 911 lines. Report-writing tools.
As neighborhoods lose actual human police presence, the industry gains automation. The decision-making process is being handed to algorithms. Tech startups are selling AI as an “automated air traffic control” system. It’s a centralized digital brain. It processes data—often collected by the very same companies selling it—and helps departments delegate resources.
Not everyone is buying it.
“A lot of it is sales gimmicks that do not actually deliver on what the promise is.”
— Abrem Ayana, Police Captain, Brookhaven, GA
Without federal oversight, cops have no choice but to trust the vendors.
Decades ago, departments used early tech like CompStat and PredPol. They promised unbiased, data-driven policing. They failed. Instead, they exacerbated racial biases. But at least humans were in the driver’s seat then. Humans could be held accountable. Humans could be wrong in understandable ways.
This new wave of AI in policing promises to fix past failures by flooding the system with real-time, objective data. Advocates say it bridges the gap. Critics say it erodes transparency when public trust is already fraying.
Drowning in Data
Jason Truppi, former FBI cybercrime agent, says cops are drowning.
He wears Meta Ray-Ban smart glasses. He talks fast. His sentences are peppered with corporate jargon. In 2020, he co-founded ForceMetrics. The company offers “Velocity,” an AI-powered decision-assist platform. It claims to turn overwhelming data into clear insights.
Truppi calls legacy police systems “antiquated.” Two decades of emergency logs, parole files, and body-cam footage have created information overload.
“We don’t use the ‘P word’ at all, because it [predictive policing] failed,” Truppi said.
Velocity is a Real-Time Crime Center (RTCC). The NYPD adopted the concept twenty years ago. Traditionally, human analysts aggregated data from 911 calls, CCTV, and license readers. They gave officers a summary before they arrived at a scene. Less surprise. Less chance of things going south—“guts and guns,” as the industry euphemism goes.
But the deluge is too big. By 2019 alone, the NYPD collected two years of body-cam footage every week. Humans can’t analyze that. Not meaningfully.
Modern RTCCs like Velocity use AI to find patterns in the chaos. The goal? Better situational awareness. Truppi blames recent tragedies in police-civilian interactions on a lack of a “data-driven approach.”
The Reality Check
Nina Loshkajian isn’t convinced.
She’s a fellow at NYU’s Center on Race, Inequality, and Law.
“The reality is police departments had already been using predictive… for years… These algorithmic systems did not prevent violent… we shouldn’t be tricked.”
Truppi’s company competes with giants like Motorola Solutions and Axon. Axon (formerly Taser) bought surveillance firm Fusus in early 2024 to launch Axon Fusus. They already make stun guns, body cameras, and license plate readers. They even have an AI chatbot and drone program.
Axon and Motorola are trying to monopolize the stack. From data collection at crime scenes to high-level strategy via AI.
Police departments often sign multiyear contracts. They get free trials. They use “sole-source procurement,” which cuts out competition. You sign once, and you’re locked in.
The Gold Rush
The money is pouring in.
In late 2023, Axon launched the “AI Era Plan.” Pay one annual fee. Get current AI tools and whatever they launch next.
Between Q1 2024 and Q1 2025, subscriptions jumped 140 percent. Axon’s President Joshua Isner told investors, “We are determining to become the AI company in public safety.” AI revenue grew 700 percent.
Andrew Guthrie Ferguson, a law professor at GWU, sees a gold rush.
“We’re seeing a gold rush… with the promise that it will make their jobs easier.”
Investors noticed. At the IACP conference, 25 percent of showroom attendees were equity firms. They weren’t cops. They were looking to bet on the tech.
And the sales pitch is working.
The Paperwork Problem
Why do cops buy?
Because they hate paperwork.
An Axon study shows officers spend 40 percent of their shift writing reports. Traffic stops. Noise complaints. Mundane stuff.
“We didn’t sign up to sit behind a key-board,” said John Mackey, a Colorado sergeant. His department uses Field Notes, an AI report tool from Truleo.
Axon’s Draft One tool promises to fix this. It’s built on a modified ChatGPT. Axon claims it’s “hallucination-free.” They turned the creativity dial down to zero.
Don’t bet the farm on that claim. Even OpenAI and Google struggle with hallucinations. Earlier this year, Draft One generated a report saying an officer had turned into a frog. Audio from The Princess and the Frog playing at the scene likely confused the AI.
It’s funny. Until it’s not.
Accountability Vanishes
Real outcomes from AI reports can be deadly.
When a human writes a report, you can cross-examine them in court. You can ask why they included specific details. You can assess their state of mind. You can scrutinize their judgment.
You cannot do that with a black-box algorithm.
Worse, Draft One originally had no audit trail. Once you submitted the report, the original AI generation was gone. No cloud copy. No version history.
Noah Spitzer-Williams, Axon’s product manager, admitted in a roundtable that the platform didn’t save originals by design. “The last thing we want is to create disclosure headaches.”
In other words, if an AI hallucination ended up in a legal case, lawyers and judges had no way to prove it was AI-generated. The human officer just signed off. It became their word against… nothing. A void.
That changed in December. After pressure, Axon updated Draft One. Now departments can retain the original AI narrative.
Brandon Garrett, a Duke Law professor, is still wary.
“The idea that you’d be making [up] data… to be used in court is dangerous.”
Generative models make things up. That’s how they work. Using that output in a judicial process is a fundamental clash with due process. The technology moves faster than the law can catch up. And while the executives are counting their 700 percent revenue growth, the people on the receiving end of these reports are left guessing what’s true, and what was dreamed up by a machine.
