Responsible Innovation
Facial Matching
A body-worn camera feature in development that helps officers identify people wanted on active warrants or reported missing. Every potential match is reviewed and verified by a human before any action is taken.
:format(webp))
PRODUCT SAFEGUARDS
PRODUCT SAFEGUARDS
Human Reviewed
A trained human reviewer verifies potential matches
Tested for Fairness
Bias evaluation in progress across diverse populations
Privacy First
Non-matches are immediately discarded and never retained
Public Visibility
Transparency dashboard provides visibility into agency use
PRODUCT SAFEGUARDS
PRODUCT SAFEGUARDS
Human Reviewed
Tested for Fairness
Privacy First
Public Visibility
:format(webp))
PRODUCT SAFEGUARDS
PRODUCT SAFEGUARDS
Human Reviewed
Tested for Fairness
Privacy First
Public Visibility
Built Responsibly
This page is for community members who want to understand how Facial Matching is designed, what guardrails are in place at this stage of development, and why we made the choices we did. We believe how we build our products matters just as much as what we build. Our Responsible Innovation Framework is where that belief becomes practice.
What is Facial Matching
Facial Matching is built on a simple principle: AI should aid officers in their investigations to help find high-priority individuals quickly, while keeping humans responsible for every match. The optional body-worn camera feature is designed to help officers identify individuals associated with active warrants for serious crimes or missing persons cases.
When activated, the system compares an image from the body-worn camera against an agency-defined watchlist and surfaces potential matches for review. Every potential match is reviewed and verified by a trained human before any action is taken. No officer is notified based on an algorithm alone.
Facial Matching is currently in field evaluation with a small number of law enforcement agencies and is not yet broadly deployed. Throughout its development, Axon's Ethics and Equity Advisory Council has provided guidance on appropriate use, accountability safeguards, and deployment considerations. The sections below outline the initial safeguards built into Facial Matching and the community input that helped shape them.
What is Facial Matching
Facial Matching is built on a simple principle: AI should aid officers in their investigations to help find high-priority individuals quickly, while keeping humans responsible for every match. The optional body-worn camera feature is designed to help officers identify individuals associated with active warrants for serious crimes or missing persons cases.
When activated, the system compares an image from the body-worn camera against an agency-defined watchlist and surfaces potential matches for review. Every potential match is reviewed and verified by a trained human before any action is taken. No officer is notified based on an algorithm alone.
Facial Matching is currently in field evaluation with a small number of law enforcement agencies and is not yet broadly deployed. Throughout its development, Axon's Ethics and Equity Advisory Council has provided guidance on appropriate use, accountability safeguards, and deployment considerations. The sections below outline the initial safeguards built into Facial Matching and the community input that helped shape them.
:format(webp))
Key safeguards
Feedback from community members, researchers, and subject-matter experts continues to shape the design principles, safeguards, and deployment guidance for Facial Matching. We evaluate these insights alongside rigorous ethical and inclusion standards to shape our product development process. The resulting initial safeguards are a meaningful reflection of that dialogue, purposefully designed to address the core issues that surface.
:format(webp))
Human Reviewed
Every potential match is reviewed by a trained human before any notification is sent to an officer. Reviewers verify the match using images alone and all match activity is recorded in an audit log that cannot be altered.
:format(webp))
Tested for Fairness
System performance is under evaluation across diverse populations. Ongoing testing is designed to identify and address demographic disparities in accuracy.
:format(webp))
Privacy First
Images that do not result in a confirmed match are immediately discarded and never logged. Transparency tools help agencies communicate how the technology is used and governed.
:format(webp))
Public Visibility
Transparency dashboard helps agencies communicate how the technology is used and governed in their communities.
Human Reviewed
WHAT WE HEARD FROM THE COMMUNITY:
A false match can result in a wrongful stop or detention. Strong safeguards are needed including both a high-confidence score from the system and a human reviewer to verify the match.
PLANNED SAFEGUARDS:
Match Review: When a potential match is generated, a human reviewer verifies the match. The reviewer sees the watchlist image and one or more images from the body-worn camera, but not the person's name, race, or offense record.
Permanent Audit Trail: All Facial Matching activity is captured in audit logs, including the time, location, and officer involved. These records cannot be altered after the fact.
High Confidence Threshold: A potential match is only surfaced to a human reviewer if the system's confidence score meets a minimum threshold, and matches that fall below it are automatically filtered out before any human review occurs.
Human Reviewed
WHAT WE HEARD FROM THE COMMUNITY:
A false match can result in a wrongful stop or detention. Strong safeguards are needed including both a high-confidence score from the system and a human reviewer to verify the match.
PLANNED SAFEGUARDS:
Match Review: When a potential match is generated, a human reviewer verifies the match. The reviewer sees the watchlist image and one or more images from the body-worn camera, but not the person's name, race, or offense record.
Permanent Audit Trail: All Facial Matching activity is captured in audit logs, including the time, location, and officer involved. These records cannot be altered after the fact.
High Confidence Threshold: A potential match is only surfaced to a human reviewer if the system's confidence score meets a minimum threshold, and matches that fall below it are automatically filtered out before any human review occurs.
:format(webp))
Tested for Fairness
WHAT WE HEARD FROM THE COMMUNITY:
Facial recognition systems have documented accuracy gaps across demographic groups. Communities want to know how well this technology performs for everyone, and expect deployments to be paused or redesigned if meaningful disparities are found.
PLANNED SAFEGUARDS:
Independent Testing: Axon is partnering with academic experts to study system performance using diverse paid actors in real-world environments. This will establish a scientific baseline to identify and evaluate demographic disparities before deployment.
Bias Evaluation: Axon is testing ways to analyze accuracy data across different skin tones. During field trials, tools such as the Monk Skin Tone scale help assess performance across diverse groups.
Tested for Fairness
WHAT WE HEARD FROM THE COMMUNITY:
Facial recognition systems have documented accuracy gaps across demographic groups. Communities want to know how well this technology performs for everyone, and expect deployments to be paused or redesigned if meaningful disparities are found.
PLANNED SAFEGUARDS:
Independent Testing: Axon is partnering with academic experts to study system performance using diverse paid actors in real-world environments. This will establish a scientific baseline to identify and evaluate demographic disparities before deployment.
Bias Evaluation: Axon is testing ways to analyze accuracy data across different skin tones. During field trials, tools such as the Monk Skin Tone scale help assess performance across diverse groups.
:format(webp))
Privacy First
WHAT WE HEARD FROM THE COMMUNITY:
Communities expect facial recognition to be accurate across all demographics, with deployments paused if disparities emerge. To protect privacy, biometric data must be processed on-device and deleted instantly if there is no match.
PLANNED SAFEGUARDS:
Immediate Deletion: If a face does not match anyone on the watchlist, or a human reviewer rejects the match, the image is immediately discarded and never logged.
Limited Retention: Biometric data is retained only for individuals who are confirmed matches against the watchlist.
Transparency Portal: Axon has invested in public-facing transparency resources, including dashboards and plain-language documentation, to help agencies provide greater visibility into AI deployments.
Watchlist Standards and Transparency: Agency watchlists are limited to individuals with active warrants for serious offenses, and the specific warrant categories used by each agency are published publicly in the Transparency Portal.
Privacy First
WHAT WE HEARD FROM THE COMMUNITY:
Communities expect facial recognition to be accurate across all demographics, with deployments paused if disparities emerge. To protect privacy, biometric data must be processed on-device and deleted instantly if there is no match.
PLANNED SAFEGUARDS:
Immediate Deletion: If a face does not match anyone on the watchlist, or a human reviewer rejects the match, the image is immediately discarded and never logged.
Limited Retention: Biometric data is retained only for individuals who are confirmed matches against the watchlist.
Transparency Portal: Axon has invested in public-facing transparency resources, including dashboards and plain-language documentation, to help agencies provide greater visibility into AI deployments.
Watchlist Standards and Transparency: Agency watchlists are limited to individuals with active warrants for serious offenses, and the specific warrant categories used by each agency are published publicly in the Transparency Portal.
:format(webp))
Public Visibility
WHAT WE HEARD FROM THE COMMUNITY:
Watchlist governance must be transparent, featuring objective criteria, a clear appeals process, and automatic removal of inactive records. Additionally, agencies should provide public dashboards and plain-language documentation so residents can track local usage.
PLANNED SAFEGUARDS:
Transparency Portal: Axon has invested in public-facing transparency resources, including dashboards and plain-language documentation, to help agencies provide greater visibility into AI deployments.
Public Visibility
WHAT WE HEARD FROM THE COMMUNITY:
Watchlist governance must be transparent, featuring objective criteria, a clear appeals process, and automatic removal of inactive records. Additionally, agencies should provide public dashboards and plain-language documentation so residents can track local usage.
PLANNED SAFEGUARDS:
Transparency Portal: Axon has invested in public-facing transparency resources, including dashboards and plain-language documentation, to help agencies provide greater visibility into AI deployments.
:format(webp))
STEP BY STEP
How Facial Matching works:
STEP BY STEP
How Facial Matching works:
STEP BY STEP
How Facial Matching works:
:format(webp))
An officer responds to a call for service or conducts their typical field operations. Their body-worn camera starts recording according to their agency's policies, just as it would in any other interaction.
FAQs
Key terms
The following definitions explain important terminology referenced throughout this page.
Body-Worn Camera
A specialized recording device mounted on an officer’s uniform to document interactions. Facial Matching capabilities are engaged when the device is actively recording or capturing images.
Watchlist
An agency-verified directory of individuals a public safety agency is actively seeking.
Human Reviewer
A qualified person (such as a field officer or a specialist in a Real Time Crime Center) responsible for verifying algorithm-generated results before any intervention occurs.
Active Warrant
A judicial directive mandated by a court that empowers law enforcement to take a specific person into custody. These orders originate from judges rather than police departments.
False Match
An instance where the technology mistakenly matches a person as a watchlist entry. This risk is why human verification is required for every potential result.
Audit Log
An unalterable, chronological record detailing every instance of system usage, including precise timestamps, geographic data, and the identifying credentials of the officer.
:format(webp))