How the technology works
Several vendors now sell tools that turn body worn camera audio into a first draft of an incident report, using the same large language model technology behind consumer chatbots. The dominant product is Axon's Draft One, launched in late 2024, which transcribes the audio captured by an officer's body camera during an incident and generates a narrative report that the officer is expected to review, edit and then submit under their own name. Axon is also the largest supplier of body worn cameras to American police departments, and has increasingly bundled its AI reporting product alongside its camera hardware and other services.
The efficiency case has real evidence behind it. Individual departments piloting the tool have reported saving between 30 and 40 per cent of the time typically spent writing a report, and Axon has said the tool has contributed to more than 100,000 incident reports and saved a combined 2.2 million minutes of officer time. For departments contending with staffing shortages, that time saving is a genuine and frequently cited justification for adoption, and by early 2026 Axon reported that revenue from its AI products had grown more than 700 per cent year on year.
Where it has gone wrong
The clearest documented failure is also among the most striking. In a case reported from Utah, background audio picked up by an officer's body camera happened to include dialogue from the film The Princess and the Frog, and the AI system incorporated it into the draft report, producing a narrative in which the officer was described as having turned into a frog. The example is almost comic, but it illustrates a serious underlying point identified by the Electronic Frontier Foundation and other critics: a system generating a police report from audio has no way of distinguishing the actual incident from anything else the microphone happened to pick up, and an officer reviewing a plausible sounding draft under time pressure may not catch every fabricated or misattributed detail before submitting it as their own account.
Public records analysis by Forbes in 2026 found documented cases of AI generated reports containing factual errors once submitted, and internal communications showing officers and supervisors expressing frustration at colleagues treating the tool as a substitute for actually writing an account of events themselves, rather than as a drafting aid to be carefully checked. One internal email described an officer's reliance on the tool to avoid writing reports at all as, in a supervisor's words, a clear example of laziness.
The transparency problem
An investigation published by the Electronic Frontier Foundation in 2025 identified a specific and consequential design choice in Axon's product: once an officer exports a finished report, Draft One deletes the original AI generated draft, along with any record of which parts of the final report were written by the AI and which were edited or added by the officer. The practical effect, EFF argued, is that if an officer's courtroom testimony later contradicts their written report, there is no way to establish whether the disputed passage was the officer's own words or an artefact of the AI draft they failed to catch, an outcome the design of the tool makes difficult to disprove either way.
A related and, in some ways, more basic problem has also emerged in practice. Oversight reviews in at least one jurisdiction found that police chiefs using Draft One did not know which of their officers' reports had actually been drafted with AI assistance, because officers were copying the AI generated narrative directly into reports submitted as their own original work, bypassing whatever disclosure markers or audit trail the system was designed to produce. Any transparency safeguard built into these tools depends on officers actually using the disclosure features as intended, which is not guaranteed simply because the feature exists.
Prosecutors and regulators are starting to respond
Some prosecutors have declined to accept the practice outright. The Prosecuting Attorney's Office in King County, Washington, told law enforcement agencies in 2024 that it would not accept reports written with the help of AI, citing exactly the kind of transparency and reliability concerns described above. Legislators have moved more cautiously but are beginning to act. Utah was the first American state to require that police disclose when a report was AI assisted. California became the second in October 2025, when Governor Gavin Newsom signed SB 524 into law, requiring AI assisted reports to be clearly marked as such, obliging agencies to maintain an audit trail identifying who used the tool and what footage informed the report, and requiring the original AI generated draft to be retained for as long as the final report itself, specifically to close the gap that critics had identified in how some tools handled draft retention.
No comparable statutory disclosure requirement yet exists in the United Kingdom, where AI assisted report drafting remains at an earlier and less widely reported stage of adoption than in the United States.
What is not settled
There is no independent, methodologically rigorous study establishing how often AI generated police reports contain material errors once submitted, as distinct from the individual documented cases and internal communications that have surfaced through public records requests and investigative reporting. Nor is it settled whether disclosure and audit trail requirements of the kind California has now legislated will be robustly enforced in practice, given that the same evasion identified in Draft One's early use, officers copying AI text into reports without flagging it as such, could in principle recur under any system that still depends on an individual officer choosing to follow the disclosure process correctly. How courts will treat a contested police report once it emerges that part of it may have been AI generated, particularly where the underlying AI draft was not retained, has also not yet been tested at scale.
Follow the coverage
PoliceAI News tracks every ai police reports story from across the English-speaking world as it breaks: new deployments, court rulings, parliamentary debates, academic research, and community responses. The feed refreshes every 30 minutes.
View Live AI Police Reports StoriesYou can also browse every ai police reports story in the feed, or explore every subject PoliceAI News covers.