The seven principles of privacy by design for physical AI

Ann Cavoukian's seven privacy-by-design principles, applied to a physical AI stack.

The seven, applied

1. Proactive, not reactive. Default to no facial recognition, no biometrics, no raw video retention.
2. Privacy as default. The edge discards frames in under a second. The event stream is counts, not footage.
3. Embedded into design. Not a config flag. The model pipeline, network stack, storage layer all assume minimisation.
4. Full functionality. Occupancy, dwell, queue and PPE are all produced from the event stream.
5. End-to-end security. Hardware encryption, signed models, audit logs.
6. Visibility. Every event we emit is documented. We will walk your auditor through the architecture.
7. Respect for the individual. No individual is tracked across cameras or over time.

Evaluation checklist

Ask any vendor the same three questions: does raw video leave my site? Can you identify an individual across cameras? Where is the data hosted? Ask for the answer in a diagram and a retention schedule, not in a deck.