Conditions, not identities
- The AI detects objects and situations: a person without a helmet, a forklift near a pedestrian, a blocked exit, a spill
- It does not use face recognition and does not match people to names, employee codes or other identities
- It does not build profiles, scores or league tables of individual workers
A person reviews every alert
- Each alert goes to a safety officer or a person nominated by the plant
- No action is taken on an alert until it has been reviewed
- AI never takes disciplinary decisions; reviewers are asked to respond by fixing conditions or training teams
- Dismissed alerts are used only to tune that plant's cameras
Where video goes
- Video stays within the plant's systems where possible, using an on-site edge device
- Only short clips or still frames linked to an alert are stored in the Service
- Continuous video is not copied to the Service
- Clips are encrypted in transit and at rest
Retention
Alert clips are kept for the period set by the plant, 90 days by default, and then deleted. Clips attached to an incident investigation or a CAPA as evidence are kept with that record.
Training the AI
Your plant's video is not used to train models for other customers without your organisation's written agreement. Feedback from your reviewers is used to adjust your own cameras' zones and sensitivity.
The plant's responsibilities
- Display CCTV notices at entrances and monitored areas as required by law
- Tell workers and their representatives what AI Hazard Eye looks for and how alerts are used
- Limit access to alert clips to people who need it for safety work
- Do not point cameras at washrooms, changing rooms, rest areas or other private spaces
Limits
AI Hazard Eye can miss events and can raise false alarms. It is an aid to supervision, not a replacement for it, and it is not a certified fire, gas or emergency detection system.