American police are grappling with a problem no one expected. The department in Heber City, Utah, began testing artificial intelligence for writing police reports, and the result? One of the documents claimed that an officer physically transformed into a frog while on duty. The department subsequently had to explain the incident publicly.
The Heber City Police Department decided to test two AI-powered software tools: Draft One from technology company Axon and a competing product called Code Four, developed by two 19-year-old MIT dropouts. Both programs work on the same principle: an officer turns on a body camera, the program captures the audio, and then automatically creates a police report from the recording. Less writing, more patrolling. At least, that is how it sounds in the marketing materials. Testing took place in December 2024.
A frog in an official document
One of the officers had a movie playing in the background while on duty. Specifically, it was Disney’s 2009 animated comedy The Princess and the Frog. The body camera recorded everything that was happening, including dialogue from the television. The artificial intelligence absorbed it all and incorporated a story into the resulting police report in which the officer turned into a frog.
Sergeant Rick Keel described it to local television station Fox 13 as follows: “The body-camera software and the report-writing software picked up the movie playing in the background. It was The Princess and the Frog. That was when we realized how important it is to review these AI reports." The department had to issue a correction and confirmed that no amphibians serve among its officers.
The incident might have been nothing more than an amusing anecdote. But it is not. It demonstrates exactly what experts have warned about ever since these tools began spreading among police forces, not only in the United States. Artificial intelligence has no common sense. It has no context. It takes sounds, words, and sentences and assembles them into text. It does not distinguish between what a suspect said, what an officer said, and what someone else said in the background—or even what happened to be playing on television. It simply processes everything and presents it as an official document.
American University law professor Andrew Ferguson warned about this as early as 2024: the automation and ease of the technology could cause officers to become less careful when writing reports. The AI report could become the default version of the truth without anyone reading it thoroughly.
AI is being deployed across America
Axon launched Draft One in early 2024. The program uses language models from OpenAI, the same technology as ChatGPT, and generates complete police reports from body-camera audio. The company claims that it cuts report-writing time roughly in half. Keel confirmed that the tool saves him six to eight hours a week. That is a tangible benefit for departments where each officer reportedly spends up to 15 hours a week on paperwork.
Axon emphasizes its safeguards: every report must be reviewed and signed by a human, and the system stores records of who generated what and from which recording. By default, the tool is limited to less serious cases. In practice, however, things look different.
The nonprofit organization Electronic Frontier Foundation published the results of its investigation in July 2025. Draft One is designed to resist audits, not facilitate them. The organization discovered that one Utah department produced more than 900 reports generated by this artificial intelligence between September 2024 and April 2025, including in cases such as kidnapping and assault. Yet the company itself says the tool should not be used for serious crimes.
The key problem: the final document often does not reveal which parts were written by the officer and which were generated by the machine. If such a file contains an error or distortion, the officer can blame the software, while the company can point to the signature of the officer responsible for reviewing the document. Mother Jones magazine also found that some departments deliberately disable the oversight tools Axon built into the system.
The future of AI in policing
This is not just about frogs and typos. Generative artificial intelligence has been shown to reproduce biases against women and people of color. When such a system writes police reports, it can subtly alter the tone of a description, choose particular words, or smooth the chaos of a real-life situation into a story that is too tidy and too convincing.
Aurelius Francisco, co-founder of the Oklahoma-based Foundation for Liberating Minds, pointed out that the fact the technology is supplied by the same company that sells Taser stun guns to police is itself worth considering.
The department’s Code Four trial period will end soon. Its leadership has not yet decided which tool, if any, it will retain. Keel suggested that artificial intelligence will remain in the department in some form. Probably not the kind that turns police officers into frogs.
Meanwhile, lawmakers in California are considering a bill known as SB 524, which would require disclosure of when and how a generative AI tool was used to write a report. If passed, it would be the first law of its kind in the United States. And the Heber City case? For now, it serves as the most famous example of what happens when technology races ahead and no one asks whether it is ready.
Sources: yahoo.com and forbes.com
