Axon Police AI: Body Cameras, Draft One, and Fusus

At a glance

Draft One makes a draft; Vision makes an alert. An officer approves a Draft One report, while a trained person decides what happens after a Vision activity alert.

Sources

1. 1. axon.com · draft one
2. 2. axon.com · auditing reporting
3. 3. axon.com · axon fusus
4. 4. axon.com · how lafayette pd used draft one
5. 5. lafayette.in.gov · 10242023 3087
6. 6. eff.org · axons draft one designed defy…
7. 7. axon.com · axon vision
8. 8. axon.com · rethinking vision building sy…
9. 9. sec.gov · axon 20260506xex991
10. 10. axon.com · prioritizing officer safety a…
11. 11. cbsnews.com · police body worn camera taser…
12. 12. baltimorebrew.com · baltimore to spend a record a…

Data behind this diagram

How Draft One and Vision pass work to people

Surface	Role in the path	Record to request
Body-worn camera	Captures video, audio, time, device, and event context	Activation policy, buffering, upload behavior, and device logs
Axon Evidence	Stores and manages digital evidence and audit trails	Retention schedule, access log, sharing log, and deletion exceptions
Transcript	Provides the text input used by Draft One	Transcript, source evidence IDs, and transcription error records
Draft One	Generates an initial report narrative from body-camera audio transcripts	Feature settings, original draft, usage audit, disclosure, and incident restrictions
Officer review	Fills missing information, edits, attests to accuracy, and approves submission	Edit history, sign-off, supervisor review, and correction records
Records system	Receives the reviewed final report	Final report, revision history, disclosure, and retention policy
Fusus	Unifies live video, alerts, locations, dispatch, ALPR, drone, and third-party feeds	Integration inventory, access controls, incident exports, and audit trail
Community Connect	Supports registration or connection of participating private cameras	Participation terms, owner consent, live-access limits, and revocation process
Approved fixed feed	Provides the agency-selected camera input for Axon Vision	Approved camera inventory, purpose, access policy, and retention
Vision activity model	Evaluates configured activity categories without documented identity recognition	Enabled categories, thresholds, model version, and false-positive audit
Lower-confidence review	Can route detections to trained human reviewers before operational notification	Reviewer contract, training, escalation rules, and review logs
Fusus alert	Presents the activity alert inside the real-time operations workflow	Alert log, recipients, response record, and Evidence transfer
Trained-person decision	Retains dispatch and enforcement authority outside the Vision model	Decision log, dispatch record, enforcement action, and policy

Scroll →

Axon, the company that sells Tasers and police body cameras, now also sells AI that writes the first draft of a police report from body-camera audio (Draft One), and software that watches live camera feeds around a city and sends officers alerts (Fusus and Vision). Police reports help decide charges and court outcomes, so who writes them, what gets kept, and who can check the record matters to anyone who is stopped, arrested, or recorded.

A body camera is no longer a stand-alone recorder. The same company's software stores the footage, transcribes it, drafts the report, and can link city and private cameras into one live view. What matters is the whole path from recording to report to courtroom, and the settings a department chooses at each step.

Draft One turns audio into an official-document workflow

Axon says body-camera audio is uploaded, transcribed, and used to generate a draft report narrative. The company describes an officer-in-the-loop process: the officer must fill missing information, edit the draft, attest to accuracy, and approve it before submission. Axon also documents audit events for draft requests, evidence IDs, users, IP addresses, incident information, and sign-off.

That workflow creates several distinct records: the original audio, transcript, generated draft, officer changes, final narrative, disclosure, usage audit, and any retained intermediate version. They should not be collapsed into “the report.” Each record can be kept for a different length of time, and each can matter differently when evidence is turned over to the defense in court.

One named deployment has a public procurement trail

Axon's July 2026 case study names Lafayette Police Department in Indiana as a user of Body 4, Fleet 3, Evidence, Draft One, Records, and Fusus. The City of Lafayette's October 24, 2023 Board of Public Works agenda separately lists a master services and purchasing agreement with Axon for the police and fire departments. Neither document shows the executed agreement, the price schedule, or the settings the department actually chose.

Original drafts are a policy and version question

EFF reported in July 2025 that Draft One's then-described workflow did not retain the generated draft or later versions, which made independent auditing difficult. Axon's current help documentation describes an optional feature for retaining and reporting original AI-generated drafts, with settings for when they are stored and deleted. Both records matter: the criticism documents the earlier transparency problem, while the current manual documents a product capability that an agency may or may not enable.

For any specific case, what counts is the software version, the setting in force on that date, and the actual audit entry, not what current documentation says.

Human review is necessary but not sufficient

Officer approval assigns responsibility, but it does not measure whether the transcript missed speech, whether the generated narrative framed an encounter differently, whether required disclosure remained enabled, or how closely supervisors reviewed the result. Those questions need paired records and outcome data.

Fusus expands the frame beyond one recorded encounter

Axon describes Fusus as a real-time operations platform. Its product page lists officer locations, dispatch data, license-plate-reader (ALPR) overlays, alerts, body cameras, drones, public-safety cameras, and participating private camera feeds in one operational view. It also describes transferring incident-linked material into Axon Evidence.

That is a different privacy boundary from a body camera recorded during one encounter. The review must include live-access rules, which cameras are connected, when a private camera can be viewed, who can search ALPR or location overlays, how an incident is created, and which material transfers into long-term evidence storage.

Axon Vision adds activity alerts inside Fusus

Axon's product page says an agency chooses the fixed-camera feeds and activity categories that Vision checks. It describes activity detection rather than identity recognition, with no facial or biometric identification. Lower-confidence detections can go to trained reviewers, and a person remains responsible for dispatch or enforcement decisions.

Axon's North Carolina corrections case study says Fusus expanded across three large prison facilities and connects that system to Vision alerts. This is Axon's account of the deployment rather than an independent statement from the corrections department.

A May 2026 shareholder letter was still targeting field trials for mid-2026 and wider availability in early Q4. In Baltimore, CBS reported approval of a ten-year Axon contract with provisional testing for AI functions, while Baltimore Brew reported that the package included Vision. The public reporting does not confirm that Baltimore is running Vision in production.

Vendor claims and external criticism

How to check your own city

Every product above leaves a paper trail. To see how your own police department uses these tools, send a public-records request to the department or city clerk for:

• The Axon contract, its price schedule, and its renewal date, which show what was bought and when it next comes up for a decision.
• The body-camera policy: when officers must record, who can view footage, and how long it is kept.
• Whether Draft One is turned on, since when, for which incident types, and whether original AI drafts are retained.
• The audit logs for AI-generated reports and any instructions prosecutors have issued about disclosing them.
• The list of cameras and feeds connected to Fusus, including private cameras, and who can view them live.
• The camera feeds and activity categories approved for Vision, and any false-positive reviews.

The data table under the figure above names the specific record to request for each step of the pipeline.

Sources

• Axon Draft One product page
• Axon Draft One auditing and reporting documentation
• Axon Fusus product page
• Axon case study: Lafayette Police Department deployment
• City of Lafayette Board of Public Works agenda: Axon master agreement
• Axon Customer Experience Improvement Program disclosure
• EFF: Draft One transparency investigation
• FTC: Axon and VieVu administrative complaint announcement
• Axon Vision product page
• Axon: Vision design and rollout statement
• Axon Q1 2026 shareholder letter filed with the SEC
• Axon case study: North Carolina corrections
• CBS Baltimore: approved Axon contract and provisional AI testing
• Baltimore Brew: Axon Vision in the proposed package