Overview#
Before body-worn camera footage can be released under a public records request, every face that does not belong to the subject of the request must be blurred. Before a recorded interview is shared with defence counsel, a witness's voice may need masking to protect their identity. Before surveillance video goes to a civil proceedings disclosure, vehicle plates require redaction. The Media Redaction Engine handles all of these scenarios automatically, applying AI-driven detection to video, images, and audio, then producing court-admissible redacted versions while preserving the unaltered originals in a separate, access-controlled evidence track.
This module is essential for law enforcement public disclosure compliance, court evidence preparation, and any intelligence context where media must be shared with parties who have limited authorisation.
Key Features#
- AI-powered face detection and blurring in video and image evidence, including tracking across frames
- Licence plate detection and redaction for vehicle privacy in surveillance and BWC footage
- Audio redaction with voice masking and segment removal to protect source identities
- PII detection in document images with automated redaction of names, addresses, and identifiers
- Real-time preview of proposed redactions before any changes are committed
- Batch redaction processing across multiple evidence files with consistent policy application
- Configurable redaction styles: blur, pixelation, solid fill, or colour replacement
- Original evidence preserved separately with redacted versions tracked independently in the chain of custody
Use Cases#
- Redacting non-subject faces in body camera footage prior to public records release, meeting FOIA and equivalent disclosure obligations
- Blurring licence plates in surveillance video to satisfy privacy compliance requirements before external sharing
- Applying audio redaction to protect witness identities in recorded interviews prepared for disclosure
- Batch-processing discovery material to redact PII before legal production to opposing counsel
- Producing redacted versions of military video for coalition partner sharing under restricted release protocols
Integration#
The redaction engine connects with evidence management, the export and sharing module, and privacy compliance workflows. All redaction actions are logged in the audit trail with operator identity, timestamp, and the specific entities redacted. Redacted versions and original files are stored in Cloudflare R2 under separate access controls, with cryptographic hashes ensuring neither copy is modified after the redaction event.
Open Standards#
- GraphQL (June 2018 Specification): all redaction operations, detection, application, batch processing, audit trail retrieval, and export, are exposed as GraphQL queries and mutations, enabling typed, introspectable access from any compliant client.
- JSON Web Token / RS256 (RFC 7519 / RFC 7517): every call to the redaction API is authenticated via an RS256-signed JWT validated against a JWKS endpoint, ensuring operator identity is cryptographically bound to each audit entry.
- W3C Verifiable Credentials Data Model v2.0: evidence items (including redacted copies) are issued Ed25519-signed Verifiable Credentials with DID-based issuer identifiers, anchoring chain of custody at collection and at each custody transfer.
- SHA-256 (FIPS 180-4): cryptographic hashes are recorded for every redacted file version and for the preserved original, allowing any consumer to verify that neither copy was altered after the redaction event.
- RFC 3161 (Time-Stamp Protocol): disclosure bundles containing redacted evidence are submitted to a trusted TSA, embedding a verifiable timestamp token (DER-encoded) in the bundle manifest for court-admissible proof of when the redaction was finalised.
- IANA Media Types (MIME): MIME type metadata on each evidence file determines which detection pipeline is invoked, face/plate detection for images and video, audio PII detection for audio recordings, and document text detection for PDFs and images of documents.
- RFC 6455 (WebSocket Protocol): real-time redaction progress and completion events are broadcast to analyst workstations over a WebSocket channel, enabling live preview updates without polling.
Last Reviewed: 2026-02-23 Last Updated: 2026-04-14