Chula Vista is joining the growing number of cities across the country using artificial intelligence to write their police reports in a move critics say will reduce transparency in their work.
The City Council voted unanimously last week to advance the purchase of the AI tools from Axon, the Arizona-based company that sells body-worn cameras and Tasers to law enforcement agencies across the country. The city will spend close to $1 million for Axon’s AI tools, including the report-writing software Draft One, over the next four years.
A group of officers had already been quietly “beta-testing” the software in the field and used it to write more than 200 reports, including investigative and arrest reports, the department said.
Chula Vista Police Department (CVPD) spokesperson Sgt. Anthony Molina told KPBS that 30 officers had tested the tool between November 2024 and early January 2025. (Voice of San Diego previously reported that roughly eight officers tested Draft One.)
CVPD leaders pitched the purchase as a time-saving move that would free up officers to handle other casework more quickly.
“Our officers are still accountable for everything they put in this police report, leading up to testimony in court,” said Dan Peak, Chula Vista’s assistant police chief, during last week’s council meeting.
Privacy and civil rights advocates, however, said Axon’s AI software raised fresh questions about the reliability of police reports, which play a central role in the criminal justice system.
“The judge or the prosecutor doesn't know which portions were written by the AI and which portions were written by the officer,” said Matthew Guariglia, a senior policy analyst with the Electronic Frontier Foundation. “It interjects a lot of uncertainty — and a lot of deniability for the officer.”
The move by Chula Vista comes as California is considering stricter regulations on how law enforcement can use AI to write police reports.
Last month, state lawmakers passed a bill that would require law enforcement agencies to clearly mark any reports produced using AI and to store all AI-generated drafts. The bill is now with Gov. Gavin Newsom, who has until mid-October to decide whether to sign or veto it.
In a statement, Axon defended the software but declined to say how many police agencies in San Diego County or the Imperial Valley were already using it.
“Axon recognizes the immense promise of ethical AI innovation and aims to harness cutting-edge AI technology to revolutionize public safety,” Axon spokesperson Kristin Lowman wrote in an email. “As with narrative reports not generated by Draft One, officers remain fully responsible for the content.”
How Draft One works
Chula Vista Police Department boasts about being among the first in the nation to adopt cutting-edge technology, such as drones and automated license plate readers.
Draft One works by analyzing the audio from an officer’s body-worn camera. The software uses a modified version of OpenAI’s ChatGPT software to transcribe the audio and produce a draft report of the interaction. The officer is then supposed to review the report, correct any mistakes and remove any sensitive information.
Finally, Draft One prompts the officer to complete an acknowledgement.
“I have reviewed the report in detail, made any necessary edits, and believe it to be an accurate representation of my recollection of the reported events,” reads the acknowledgement shared with the City Council. “If needed, I am willing to testify to the accuracy of this report.”
After that, the officer creates a copy of the draft and the original AI-generated text disappears.
Privacy advocates see tradeoffs
In a July presentation for Chula Vista’s Privacy Protection and Technology Advisory Commission, Sgt. Pricilla Graton said a “select group” of officers had already been testing Draft One to write reports as part of a pilot program.
Some officers, Graton said, reported being able to complete reports “hours faster” than their colleagues.
But privacy advocates say there’s a worrying tradeoff for that supposed efficiency.
Guariglia fears Axon has made the software too customizable, making it simple for agencies or officers to override the program’s safeguards. For example, agencies can disable the default header or footer, which clearly labels reports that have been produced using Draft One.
Lowman, with Axon, said the software included safeguards that agencies cannot turn off, including “mandatory officer sign-off, required placeholder statements, and a secure, logged audit trail.”
The fact that Draft One deletes its initial AI-generated text is another concern for Guariglia. Without the original draft, he said, the use of AI could make it easier for officers to get away with lying in their reports.
“They can just say, ‘Actually, that part in the report, the AI wrote that part,’” Guariglia said. “The consequences for an officer not checking the AI's work well enough are quite different than the consequences for if they are caught lying in their police report.”
The American Civil Liberties Union has also criticized these AI tools. In a report last November, the organization said the technology is “too untested, too unreliable, too opaque, and too biased” to be used in criminal justice work.
Sgt. Graton acknowledged the software is not a perfect tool at the July privacy commission hearing.
She noted officers found that Draft One tended to “oversimplify the narrative” during lengthy interviews, and that they would have to stop and restart their recordings to get more detailed drafts.
Local officials on board
Those concerns did not appear to worry members of the City Council, most of whom applauded the police department’s move to embrace generative AI.
“I think we need to take stock of the fact that this technology is going to assist us in making more policing resources available,” Councilmember Jose Preciado said. “That's something we all need to support.”
The only elected official to raise concerns about Draft One was Councilmember Cesar Fernandez, who thanked City Manager Maria Kachadoorian for asking police to present their proposal to the city’s privacy commission first.
“I just believe that any new surveillance technology should be vetted and then discussed,” Fernandez said.
Chula Vista police officials have said repeatedly that San Diego County District Attorney Summer Stephan’s office has signed off on the department’s use of Draft One.
Tanya Sierra, a spokesperson for Stephan, told KPBS that the District Attorney's Office had no formal policy on police agencies using AI to transcribe body camera video.
“We will continue to evaluate cases submitted to us for prosecution based upon the standard of proof beyond a reasonable doubt,” Sierra said in a statement. “We expect police departments to submit accurate police reports.”
At least two states have already moved to regulate the use of programs like Draft One.
Earlier this year, Utah passed a law requiring law enforcement agencies to label reports written using AI and attest to their accuracy.
California’s bill would impose additional restrictions, including making it easier for independent auditors to evaluate AI tools and barring tech companies from selling data from body-worn camera footage.
