Law enforcement agencies, frequently among the first to adopt cutting-edge technology, have once again embraced artificial intelligence as a potential breakthrough. After successfully integrating AI-powered audio transcription tools, departments moved on to a far more ambitious application: software that drafts entire police reports. In one early demonstration, an Oklahoma City sergeant who had spent nearly an hour searching for suspects with his K-9 let AI write the first draft of his report — and the tool produced it in about eight seconds, pulling from the audio and radio chatter captured by his body camera.
Draft One is the generative AI tool that Axon — the company best known for its Tasers and body cameras — unveiled in April 2024 to speed up the laborious process of writing reports. Built on Microsoft’s Azure OpenAI platform, it analyzes the audio from an officer’s body camera and generates a draft narrative in minutes. According to The Associated Press, the software promises to trim paperwork by up to an hour a day, freeing officers to focus on mental health calls and community engagement.
Whatever its advantages, the technology has raised pointed questions about reliability and bias. Large language models like the one underpinning Draft One — the same class of technology as ChatGPT — have been criticized for their tendency to produce false or misleading output. Even though Axon says it has tuned its model to minimize speculation and embellishment, errors and so-called hallucinations remain possible. The company also revealed that it experimented with computer vision to summarize what a body camera “sees,” but pulled back, with CEO Rick Smith acknowledging that, given the sensitivities around race and identity in policing, the feature needed far more work before release.
There are also legitimate concerns about gender and racial bias creeping into AI-generated reports. Researchers have repeatedly demonstrated such biases in large language models, and deploying them in law enforcement could deepen inequalities that already exist in the system.
Departments use Draft One in different ways. Some agencies have given officers broad access to use it on any report, while others restrict it to minor incidents; Axon itself initially limited new clients to lower-stakes cases that do not involve felonies or arrests, so agencies could grow comfortable with the tool before expanding its use. But experts argue that leaning on AI alone is not a viable option, given how damaging an error in a police report can be.
“The large language models underpinning tools like ChatGPT are not designed to generate truth. Rather, they string together plausible sounding sentences based on prediction algorithms,” Purdue clinical associate professor Lindsey Weinberg, who focuses on digital and technological ethics, told Popular Science.
Weinberg, who directs the Tech Justice Lab, argues that “almost every algorithmic tool you can think of has been shown time and again to reproduce and amplify existing forms of racial injustice.” Researchers have documented numerous instances of gender and racial bias in large language models over the years.
“The use of tools that make it ‘easier’ to generate police reports in the context of a legal system that currently supports and sanctions the mass incarceration of marginalized populations should be deeply concerning to those who care about privacy, civil rights, and justice,” Weinberg concluded.
The question posed in this article’s headline — whether AI-drafted reports hold up in court — has only grown sharper as the technology has spread. Draft One has become Axon’s fastest-growing product, adopted by departments in cities such as Fort Collins, Lafayette, Tampa, and Campbell, and it is no longer alone in the market, with rival firm Truleo offering a comparable tool called Field Notes. Because police reports sit at the heart of the criminal justice process — shaping whether prosecutors bring charges, whether judges grant bond, and how officers later testify — the stakes of an automated first draft are unusually high, as American University law professor Andrew Guthrie Ferguson and other scholars have noted.
Prosecutors have already begun pushing back. In September 2024, the King County Prosecuting Attorney’s Office in Washington state, which includes Seattle, instructed police not to submit reports drafted with AI, warning that the tool would “likely result in many of your officers approving Axon drafted narratives with unintentional errors in them.” Civil-liberties groups have echoed the concern: the ACLU published an analysis recommending against Draft One, and the Electronic Frontier Foundation has warned that the product is, by design, difficult to audit — because once an officer exports a finished report, Draft One does not retain the original AI-generated draft or a log of which sentences came from the machine versus the officer. Critics warn that gap could let an officer whose courtroom testimony contradicts their report simply blame the AI.
States have started to respond. In 2025, Utah became the first state to require police to disclose when a report was written with AI and to certify that an officer reviewed it for accuracy, while also mandating that agencies adopt formal AI policies. In October 2025, California followed with a broader law that requires disclosure of AI use on the report, retention of the first draft for as long as the final report is kept, identification of the report’s author, and a ban on vendors selling or sharing the data police feed into the system — provisions that directly challenge Draft One’s no-retention design.
For its part, Axon points to safeguards it says keep humans firmly in control: every report must be reviewed and approved by the officer, then run through supervisor and records review, and officers remain fully responsible for what they sign and must be prepared to testify to its accuracy. The company also cites a double-blind study involving two dozen experts from law enforcement and the courts that, it says, found Draft One reports equal to or better than those written entirely by officers on measures like completeness, neutrality, and coherence.
The growing use of AI in law enforcement demands a careful weighing of benefits against risks. The technology can boost productivity and streamline tedious procedures, but it must be deployed cautiously so that efficiency does not come at the expense of accountability, justice, or accuracy.
