By Thaddeus Miller
The Fresno Bee
FRESNO, Calif. — For almost a year, the Fresno Police Department has been deploying artificial intelligence to write many of their tens of thousands of police reports.
Fresno is believed to be one of the first and largest U.S. cities to deploy what’s known as generative AI to write police reports informed by officers’ body-worn cameras. But, while officials say AI has made the often time-consuming practice of writing police reports more efficient, the technology’s use has drawn concerns from civil rights groups and defense attorneys.
Officers have been outfitted with body-worn cameras made by Axon Enterprises since 2015. Last April, the department began using Axon’s First Draft program to start writing police reports.
The company and Fresno police say the program is meant to be a time saver in the more than 90,000 reports written by Fresno officers in a year. Neither police nor Axon provided proof the cameras save time.
Fresno Chief Mindy Casto said the department does not have a scientific way to calculate time saved, but has estimated a 20% savings.
Officials with Axon said departments average a savings of about 23 minutes per police report.
Casto said the department has rolled the program out slowly, only using the software on misdemeanors, excluding more sensitive crimes like domestic violence or resisting arrest.
“I think the value is only going to grow as we expand the program, but I want to be very cautious in the roll out to make sure we’ve mastered it before we start delving into the more serious crimes,” Casto said.
Those working outside of law enforcement say they’re concerned adding AI into the mix could violate the rights of citizens and complicate holding officers accountable.
How AI police reports work
The body cameras officers turn on during calls for service immediately send the information collected to the police department’s cloud — remote servers that collect large amounts of information.
Axon’s software then writes a “first draft” of the encounter into a format that fits police reports within about three to five minutes after an officer ends the recording, according to Noah Spitzer-Williams , Axon’s senior principal product manager of generative AI.
The AI is not meant to do all of the work, he said. Each officer is supposed to review the report, add details and respond to prompts during the process.
“At the end of the day, the officer owns the final words of the report, but draft one is here to kind of give you a good head start,” he said. “They make their edits, they sign off on it, and they’re back out in the field.”
Some officers may even dictate details they are seeing during the call so that the AI can include them in the first draft, Spitzer-Williams said.
Fresno’s relationship with Axon
Axon has had a technology-testing relationship with Fresno Police Department in exchange for steep discounts. Fresno became a “development partner” with Axon in 2017 and has access to its new technologies for “little to no cost to the city,” according to city records.
Former Police Chief Paco Balderrama discussed the Axon partnership during the June 2024 city budget hearings.
“We’re very proud of that because we have been cutting-edge as far as technology, as far as tasers, body-worn cameras, drones,” he said. “And we’ve even helped developed a lot of different products, three that I can think of right now, that we are one of the very first in the world to utilize.”
He said the department gets “about a $19 million break” compared to agencies that do not have a development partnership.
The Fresno City Council approved a five-year $18.3 million contract with Axon last year, which included about $1.3 million for the First Draft program.
Axon representatives declined to say whether Fresno is the largest city using the AI reports, citing privacy concerns.
Some other cities known to be using the technology include California towns like San Mateo, East Palo Alto and Campbell, all of which are smaller than Fresno. Other cities include Fort Collins, Colorado; and Lafayette, Louisiana.
Potential pitfalls and concerns
Experts, defense attorneys and civil rights groups remain concerned about how AI reports are used and what that could mean for the civil rights of the accused.
The American Civil Liberties Union has come out against AI reports, citing the potential for warping the memories of officers, a potential for bias AI has shown and transparency, among other reasons.
The King County Prosecuting Attorney’s Office in Washington was concerned enough to send a memo to police departments in September telling officers to not use AI in reports, citing the potential for errors.
Defense attorney Jeffery Hammerschmidt of Fresno said he has not seen an AI police report so far in court, but says attorneys often use AI themselves. But, as police should, attorneys must be diligent in double-checking the work.
“It can be a good tool but it can’t substitute the officer’s work,” he said. “If an officer was using it as a shortcut, I’m going to find those errors and use it to my advantage to find the truth.”
As a private defense attorney, he said he reviews reports and body-worn camera footage often to find inconsistencies. That could be harder on the Fresno County Public Defender’s Office.
“It’s very different for the public defender’s office with their caseload to thoroughly review all the camera footage,” he said.
Requests by The Bee to speak about the topic went unanswered by the Public Defender’s Office.
An AI startup called JusticeText is also developing the use of AI in reviewing body-camera footage, which public defenders in places like Sacramento and Modesto have used.
Axon’s program uses ChatGPT, but Axon’s representatives say their version is stripped of the kind of creativity shown in the chatbot version available online.
It is not supposed to come to any conclusions, which are called “hallucinations” when they are inaccurate. That potential for hallucinations is a concern among skeptics.
The new technology has the potential to affect civil liberties and hurt the public trust, and that requires officers to be diligent and not take shortcuts, according to Paul D. Knothe, a partner with Liebert Cassidy Whitmore in Los Angeles. The firm advises police agencies across the state.
“The computer is trained on data generated by humans, so it will reflect our biases and, in some cases, magnify the biases,” he said. “So that’s why a human check on it is critically important.”
He said many agencies would benefit from requiring police officers to save the first draft spit out by the program for comparison against the final draft submitted. That would show errors introduced by the program and, at the same time, prevent an officer who might try to blame their errant actions on the AI.
Casto, who was in an interim position until being announced as full-time chief on Feb. 13 , has said she wants to improve Fresno officers’ response to non-emergency calls.
“I think for any chief, (AI) is a fantastic tool, especially in a very large, busy agency,” she said. “Here where we’re just constantly under the gun on calls for service, it could be a game changer for the kind of service that we deliver.”
—
©2025 The Fresno Bee. Visit fresnobee.com. Distributed by Tribune Content Agency, LLC.