Redaction

Overview#

A prosecutor preparing discovery materials for a high-profile trial needs to redact the faces and badge numbers of undercover officers from several hours of body camera footage before handing recordings to the defence. Her team runs the AI auto-detection pass to blur faces and licence plates across the entire batch, reviews the confidence-flagged regions in preview mode, and then applies the final redactions with a single operation. Every step, including the initial detection run, the dry-run preview, and the final apply, is written to an immutable audit trail. If a redaction needs to be revised, the original file is preserved and a new version is created. The Redaction domain manages this entire process: AI-assisted detection, manual selection, version control, and chain of custody documentation from start to court submission.

Key Features#

Manual redaction with rectangular coordinates and polygon point selection
AI-powered auto-detection for faces, licence plates, PII, and custom patterns
Configurable confidence thresholds with dry-run preview mode
Version control preserving original files with per-redaction versioning
Immutable audit trail logging all apply, revert, view, and export actions
Batch processing for bulk redaction across multiple evidence items with progress tracking
Redaction revert and undo capabilities with original content restoration
Export generation with optional audit trail and metadata inclusion
Multiple redaction types: face, licence plate, text, audio, and custom
Error isolation in batch operations preventing single failures from stopping the batch

Use Cases#

Redaction requirements arise wherever protected identities or sensitive information must be stripped from evidence before disclosure. Key industries include law enforcement and criminal justice, legal services, and healthcare.

Redacting witness identities and sensitive PII from evidence before legal disclosure
Running AI auto-detection on body camera footage to blur non-subject faces and licence plates
Batch-processing discovery response documents with tracked redaction and audit trails
Exporting redacted evidence with chain of custody documentation for court submission

Integration#

The Redaction domain connects with evidence management, legal disclosure packages, system-wide audit logging, and chain of custody records.

Open Standards#

GraphQL (June 2018 specification): All redaction operations, including apply, revert, batch creation, and audit-trail retrieval, are exposed exclusively through a typed GraphQL API using queries and mutations.
JSON Web Token (RFC 7519) with RS256 / JWKS (RFC 7517): Every GraphQL operation requires a caller-supplied JWT validated against the platform JWKS endpoint; unsigned or algorithm-downgraded tokens are rejected with HTTP 401.
JSON (RFC 8259): Redaction coordinates, polygon points, audit-entry details, and export manifests are all serialised and persisted as JSON documents.
RFC 4122 UUID: All primary keys (redaction id, batch id, audit entry id, export id) are version-4 UUIDs, enforced at insert time with fallback generation when legacy prefixed strings are supplied.
ISO 8601 (datetime): All timestamps, including applied_at, created_at, completed_at, and export expires_at, are stored and transmitted as ISO 8601 strings; the service parses them via datetime.fromisoformat and serialises via .isoformat().
SHA-256: Each redacted-file version records a SHA-256 hash in the file_hash field to support integrity verification and chain-of-custody documentation.
IANA Media Types (MIME): The mime_type field on every redaction version record uses IANA-registered media type strings, defaulting to application/octet-stream, to identify the format of stored evidence files.
GDPR (EU 2016/679) legal basis: The legal_basis column is persisted on every redaction row, supporting documentation of the lawful ground (e.g. Article 17 right to erasure, Article 5 data minimisation) under which PII was redacted from evidence before disclosure.

Last Reviewed: 2026-02-05 Last Updated: 2026-04-14