Axon Says AI Police Reports Save Time. Public Records Show They Get Facts Wrong

Axon's AI-powered police report tool, Draft One, has been used to draft over 600,000 police reports across the United States, but public records obtained by Forbes reveal that the system frequently introduces factual errors that officers must spend valuable time correcting. The promise of saving hours of paperwork is undercut by inaccuracies—ranging from misidentified objects to wrong suspect descriptions—that undermine the technology's reliability and raise concerns about its impact on criminal justice.

Axon, the company best known for its Tasers and body cameras, launched Draft One in 2023 as a generative AI tool that automatically transcribes and summarizes body camera footage into structured police reports. The tool was hailed as a breakthrough for officer efficiency, allowing cops to complete paperwork in minutes rather than hours. However, a Forbes investigation sourced from thousands of internal police records and public data requests shows that Draft One's output is far from flawless. Officers have reported finding errors in nearly every field: dates, locations, witness statements, and even the nature of the incident itself.

One key detail from the records: a report drafted by AI stated that a suspect was armed with a "knife" when the footage clearly showed a screwdriver. Another error involved an AI report claiming an officer ran a red light, while the video evidence showed the officer had a green light. These mistakes aren't just embarrassing—they can affect criminal cases, officer discipline, and public trust. Axon claims an accuracy rate above 90%, but the real-world error rate, as measured by required human edits, appears to be significantly higher.

Cops are spending an average of 5 to 15 minutes correcting each AI-generated report, according to the data obtained by Forbes. That's less time than writing from scratch, but still far from the frictionless promise. Some departments have reported that newer officers rely too heavily on the AI and fail to catch mistakes, while others have disabled certain features due to persistent inaccuracies. Axon has acknowledged the issues and says it is working on improvements, including better contextual understanding and human-in-the-loop verification.

The broader implication is a cautionary tale for deploying generative AI in high-stakes public sector roles. Police reports are legal documents used in court, investigations, and oversight. If AI introduces errors, it could lead to wrongful arrests, flawed evidence, and erosion of public confidence. Legal experts interviewed by Forbes warn that without rigorous validation, such tools could create liability for both law enforcement agencies and technology vendors.

What happens next? Axon is expected to release an updated version of Draft One by early 2027, incorporating feedback from over 200 police departments. Meanwhile, privacy advocates are calling for independent audits. The race to automate police work continues, but the gap between marketing claims and operational reality remains wide.