Licensed, real-world. Not synthetic. Not scraped.

Every clip in our pipeline is captured through explicit, informed consent. We do not use synthetic generation, web scraping, or unlicensed footage. All data is original, real-world, and captured under a controlled protocol — so the provenance chain starts clean.

Our data operations are designed for global use. Rights clearance is multi-jurisdiction — not tied to a single country's framework. GDPR principles are applied as a baseline: purpose limitation, data minimisation, subject rights. Customers deploying in regulated markets can request documentation.

Collectors consent explicitly before any capture session. Consent covers the specific task type, use of the footage for AI training, and rights transfer. Collectors can withdraw consent and request deletion of their data. Personal data is separated from training data and handled under strict access controls.

Every dataset includes a provenance record: capture date, environment category, collector consent reference, annotation pipeline version, and QA score. This record travels with the dataset and is inspection-ready. Customers can trace any clip back to its capture context without exposing collector identity.

Data is stored encrypted at rest and in transit. Access to raw capture data is role-restricted; customer datasets are isolated and not shared between customers. Infrastructure runs on AWS (eu-west-1) with CloudTrail audit logging and KMS key management.

Egocentric captures may incidentally capture bystander faces, licence plates, or other personal identifiers. Our pipeline applies automated detection and masking for faces and plates before any data leaves the annotation stage. Additional anonymization passes can be applied on request for sensitive deployment contexts.