These tools work from the audio track of a body-worn camera recording, not the visual footage, transcribing what was said during an incident and using a generative language model to produce a structured first draft of the resulting report. Axon's Draft One is built on OpenAI's GPT models; Truleo's Field Notes is built on Amazon's Bedrock platform. In both cases the officer is expected to review, correct and sign the draft, and remains responsible for the final submitted document.

This is a narrower, more specific tool than a general-purpose assistant like Microsoft Copilot applied to policing tasks, which the site tracks separately. Report-writing tools are purpose-built for one task with a fixed input and a fixed output, rather than a general chat interface an officer might use for anything.

Scale as of mid-2026: Axon says its AI tools have helped write around 600,000 reports across roughly 600 US police departments.

Regulation has begun to arrive, and unusually it arrived quickly. Utah Code 53-25-902, effective 7 May 2025, requires every law enforcement agency to hold a generative AI policy and requires the report author to certify they read and reviewed the record for accuracy. California's S.B. 524, signed in 2025, requires disclosure on reports authored fully or partly by AI, requires retention of the first draft, and bars vendors from selling data agencies fed to the AI.

That retention requirement collides with how the product is built. The Electronic Frontier Foundation reports that Draft One by design keeps no edit log or draft history, meaning compliance with California-style rules would require Axon to change the product rather than agencies to change their practice.

A related product worth distinguishing: Axon Form One auto-fills report administrative fields from body-camera audio and has been tested by Lafayette Police Department in Louisiana, where records show it mis-transcribed names and licence plates. It sits in the same family as Draft One but performs a narrower task.