Investigation
Bend Privacy Alliance / Axon patents series, Part 9 of 11 · September 2026
Axon’s patents describe a system built to track more than video.
A camera file can be only one layer of the record. The harder question is whether a later reviewer can reconstruct how police information was created.
A police body-camera recording looks like a file. It has a beginning, an end, a timestamp and a picture. Someone uploads it. Someone watches it. Eventually it may become evidence in a criminal case.
But the video may be only one part of the record.
Axon’s patents describe cameras and evidence systems capable of preserving another layer of information around a recording: when the camera started and stopped, what triggered it, whether nearby cameras were recording, which incident the footage became associated with, whether the system expected evidence that never arrived, and what machine-generated information was later derived from the source.
Some of that architecture is no longer merely described in patents. Current Axon Evidence documentation describes immutable evidence audit events, while current product guides document source information, a Document Checksum, original-versus-derived relationships, automatic activation reasons and synchronized multi-camera review.
The result is a larger question for digital evidence. For decades, chain of custody has largely meant being able to show where a piece of evidence came from, who handled it and whether it was altered. As police information becomes increasingly automated, connected and machine-generated, that definition may no longer be enough.
The new question is whether we can reconstruct how the information itself was created.
The video is not the whole record
Consider a body-camera recording. What appears on screen tells us what the camera saw and heard. It does not necessarily tell us what the camera itself was doing.
An early Axon audit-trail patent describes a camera with a store for video and a separate store for auditable device events. Those events can include recording start, stop and pause; button presses; battery and storage problems; docking and wireless interactions; motion; geofence events; drops and other device activity. The patent family also describes digitally signing audit entries.
That creates two different kinds of evidence: the recording itself, and a machine-generated history of the device that made it.
Current Axon products preserve a meaningful portion of that distinction. Axon’s evidence system records interactions such as viewing, updating, downloading, sharing and changing evidence metadata, while device audit trails can be exported separately.
That can matter when the disputed question is not what appears in a video, but what happened around it. Did recording begin when policy required it? Was a camera docked? Did its state change? Was evidence subsequently edited, shared or placed into a case?
A video cannot necessarily answer those questions by itself. The surrounding records may be part of the evidence too.
Before recording, there can already be video
That creates another provenance question. A later reviewer may need to distinguish among what the device was merely holding temporarily, when the evidentiary event formally began, which pre-event material crossed into permanent storage, and whether the system preserves enough recording-state or audit information to reconstruct that transition.
The old Looxcie patent should not be read as a blueprint for every current Axon buffering implementation. Its significance here is narrower: pre-event memory and later preservation are part of the technical lineage of wearable-camera evidence systems.
Current Axon Body 4 documentation makes that distinction explicit. In Ready mode, the camera is capturing video in a pre-event buffer but is not recording it to permanent memory. When Recording mode is activated, the configured portion of video immediately preceding activation is saved and attached to the event in permanent memory. Axon’s Body 4 settings documentation separately describes the pre-event-buffer setting as determining the buffer duration included in the final video.
That history matters because it separates three events that can easily be collapsed in ordinary language: the camera was temporarily retaining video, formal recording began, and some of the earlier buffered material became durable evidence.
A Looxcie-origin patent, US8237856B2, “Timeshifting video recording camera”, describes a camera that continuously writes video into a circular buffer. As newer frames arrive, older material can be overwritten unless a portion of the buffer is partitioned and preserved so it is no longer lost when the temporary memory rolls forward. The patent originated with Looxcie; the company later became Vidcie, and the patent was subsequently assigned to Axon Enterprise.
The distinction between buffering and recording has a longer history in wearable-camera technology than Axon’s current body-camera line.
The system can know that evidence should exist
One of the most consequential findings in Axon’s patent portfolio concerns evidence that is missing.
Axon’s evidence-recovery patent describes urgent audit metadata reaching a server before the complete audiovisual evidence. The patent says that this early audit tag can give the server an inventory of recorded data whose later upload should be expected and verified; the broader workflow also describes pulling additional data from a recording device when needed.
That changes what “missing video” can mean. There is a large difference between no recording ever being created and a system having reason to expect evidence that never arrived.
What remains unresolved is whether Bend Police can see or export an equivalent record today. The current public Axon Evidence documentation reviewed for this project does not clearly expose a user-facing “expected but missing evidence” record.
That is precisely why this becomes a records question rather than a conclusion. If Axon systems used in Bend maintain expected-evidence records, upload failures, recovery requests or similar information, those records could matter when footage that should exist cannot be located.
Why did the camera start recording?
A second group of Axon patents deals with something equally fundamental: causation.
Axon’s distributed-control patent describes notifications from connected devices changing a camera’s recording state. The examples include vehicle light bars, conducted electrical weapons and holster sensors, and the patent allows notifications to be relayed through intermediate devices.
In that architecture, the camera does not merely know that recording began. Information about the event that caused recording can become part of the camera’s audit trail or media metadata.
This is one area where current product documentation makes the connection concrete. Axon’s Body 4 Nearby BWC Activation documentation says one nearby camera can cause another to begin recording and assigns the resulting evidence an activation reason of “Nearby BWC Notification.” Other Axon Signal workflows can also associate automatic activation with configured inputs and events.
The evidentiary question therefore becomes more precise. If a recording begins automatically, can a later reviewer determine what caused it? If it does not begin when expected, can the agency reconstruct whether the problem was the originating event, pairing, notification, configuration or camera response?
Preserving why a recording began can be as important as preserving the recording itself.
One incident can have many camera histories
Police encounters increasingly involve more than one recording device. There may be several body-worn cameras, a Fleet camera, fixed cameras or other systems capturing different portions of the same incident.
Axon’s multi-recorder patent family describes one way to turn those separate recordings into a common evidentiary event. Devices can exchange alignment information that helps relate recordings even when their clocks disagree. More strikingly, the architecture distinguishes states such as recording, buffering and not recording. As with the pre-event distinction described earlier, buffering is not simply another word for a completed evidentiary recording.
A device can potentially retain information about its own state and the states of nearby devices. That creates a broader conception of evidence completeness.
Instead of asking only whether one officer’s video survived, the question can become: Which recording devices were present, and what was each one doing?
Axon’s current Multicam documentation says recordings from Axon cameras can be reviewed together when the cameras were within roughly 30 feet of one another and recorded for at least one minute, excluding pre-event buffering. The workspace synchronizes those recordings for review. What the public documentation still does not expose is the lower-level history described in the patent: raw alignment beacons, nearby-device status history, or a recording/buffering/not-recording timeline.
Whether Bend retains any of those lower-level records is a local deployment question. The patent nevertheless shows that incident-level completeness has been contemplated in far greater detail than the ordinary public discussion of body-camera footage suggests.
The evidence an investigator watches may not be the original
Not all police video begins with a police camera. Businesses, homeowners, public agencies and other organizations routinely produce footage that later becomes evidence.
That creates a different chain-of-custody problem because third-party surveillance systems may package multiple camera channels, timestamps, audio and metadata inside proprietary formats.
Axon’s US11825156B1 patent addresses this problem directly. It describes ingesting a multiplexed surveillance file, determining how the proprietary format stores its information, and producing separate time-stamped streams for the cameras contained inside it.
The resulting playable footage is therefore a derived artifact. There is the original proprietary source file, the process used to interpret it, and the decoded video presented for analysis.
Current Axon Evidence documentation makes that lineage unusually explicit. Axon’s Evidence Extractions documentation says an extraction creates a new evidence item, never modifies the original, keeps the parent and derivative linked, and gives the extracted item its own audit trail. Axon’s Community Request evidence-protection guide separately says that proprietary third-party video conversion retains both the original file and the playable version.
That is strong ordinary provenance. But it raises another level of questions. If software converted the evidence, can a later reviewer determine which decoder or format definition was used? Were all channels extracted? Did all original metadata survive? Can the conversion be independently reproduced?
The integrity of the original file is one problem. The integrity of the process that interprets it is another.
Then machines begin adding information
Transformation does not stop with video conversion.
Another Axon patent family describes analyzing captured evidence and creating a separate layer of what the patent calls “identified data.” That information can represent objects, people, speech, events, characteristics, patterns or analytical results.
The architecture also describes alignment information connecting machine-derived findings back to their source. Depending on the evidence, that relationship can identify a timestamp, time range, frame, region within an image or multiple portions of source material.
That distinction sounds technical, but its importance is simple. There are two very different transparency models:
The software says X.
versus
The software says X, and here are the precise portions of the original evidence from which X was derived.
The second model gives a human reviewer something to test.
The same basic question now extends well beyond object detection. Axon’s current ecosystem includes transcription, automated evidence tagging, ALPR, AI analysis and AI-assisted report writing. Its July 2026 Evidence release notes describe Lead Lock analyzing evidence relationships, while Draft One documentation describes an AI-assisted reporting workflow built from police evidence.
That patent should not be read as a schematic for every newer Axon AI product. Its narrower significance is still important: Axon has patented an evidence architecture in which machine-derived information can remain tied to precise locations in the underlying source.
The biggest transparency gap is computational provenance
This research produced a somewhat unexpected result.
Axon publicly documents substantial provenance for the custody and lifecycle of ordinary evidence. Evidence audit events are immutable. Evidence Details exposes source information, Evidence Origin and a Document Checksum. Axon distinguishes originals from derivatives and maintains relationships between them, while its product documentation also records retention, sharing and other evidence-history events.
Those are meaningful safeguards.
The larger unresolved question is computational provenance: whether equally complete traceability survives when software creates new information from the evidence.
When software turns source evidence into a new police assertion, can someone later reconstruct the complete path? What model or version performed the analysis? What exact source material supported the result? Was there a confidence value? Was the output regenerated? Was it changed by a person or another machine? Did those relationships survive when the information was exported into another system or an official report?
Public product documentation reviewed for this project does not establish a single, universal and exportable Axon Evidence record containing that full history across machine-generated information.
A system can have an excellent audit trail showing that an officer opened a file at a particular time and still provide an incomplete explanation of how an AI-generated statement inside the resulting police record came into existence.
The first question concerns custody. The second concerns provenance.
Modern evidence increasingly requires both.
Bend already sits inside part of this architecture
This is not purely hypothetical for Bend.
A City software inventory produced through public records lists multiple Axon products as Bend Police Department production systems, including Axon Evidence, Evidence Sync, Evidence Upload XT, Respond Plus, Standards, Capture and View.
Separate City records document Bend’s body-camera and evidence-storage purchases, Fleet cameras, Fusus, Axon Air and a later Officer Safety Plan bundle that consolidated much of the Axon environment. Bend has also begun using Draft One, although the public procurement trail for that capability remains incomplete.
A patented capability, a commercially documented feature, a purchased product and a locally enabled function are four different propositions. Bend’s records establish the local product environment; they do not by themselves establish that every capability discussed above is enabled here.
They establish something narrower and more useful: Bend operates inside an Axon evidence ecosystem in which questions about auditability, transformation and machine-generated records are no longer abstract.
The next layer of transparency has to come from local configuration and local records.
What Bend should be able to show
The public does not need speculative claims about what a patent might someday permit. It needs records showing what the systems being used here actually preserve.
For Axon Evidence and related systems, Bend should be able to explain what its evidence and device audit trails contain; whether the platform records evidence that was expected but never received; what upload failures and recovery events are retained; whether automatic camera-activation reasons survive; what Multicam or related recording-state information is available; and which provenance fields accompany transcripts, AI tags, ALPR outputs, Draft One records and other machine-derived information.
It should also be possible to answer what happens when those records leave the police department. What does a prosecutor receive? What does defense counsel receive? Do they receive only the playable video and final report, or the audit and provenance information necessary to test how those records were produced?
Those are not demands for exotic new technology. In several instances, Axon’s own patents describe much richer provenance than a bare evidence file, and current Axon products already expose significant portions of the ordinary chain-of-custody layer.
The question is which portions survive all the way to accountability.
Chain of custody is becoming a chain of information
A camera records. A device creates an audit trail. Another system associates the recording with an incident. Software may transcribe the audio, identify something in the evidence, create a derivative file or turn portions of the source material into an official record.
Every step can create another layer between what originally happened and what a later reviewer sees.
Axon’s own patents show that the company has spent years designing ways to preserve many of those relationships: what a device was doing, what triggered it, which cameras were nearby, whether evidence was expected, where a derivative came from, and where machine-generated information appeared in the original source.
That changes what we should expect from digital chain of custody.
It is no longer enough to establish that nobody altered the final file.
We should be able to reconstruct how the information inside it came to exist.