AI transparency statement

This statement explains, in plain terms, how artificial intelligence is used in the service Mitigate It supplies, who oversees it, what it is not designed to do, and its limitations. It is published as a matter of good governance rather than legal obligation, and it follows the same rule as the rest of this site: we describe what is actually there.

What the AI does

The service applies video analytics to CCTV cameras that a business already operates. Software watches the live feed and recognises defined risk events, such as missing personal protective equipment, a person working near an unprotected edge, a moving vehicle close to a pedestrian, smoke, escape of water, or out-of-hours intrusion. When an event is recognised, an alert is sent to the people responsible for the site so they can act. Where the customer has given explicit consent, structured risk data derived from those detections can also be shared with their insurer.

The detection engine is developed by our technology partner, a specialist video analytics company whose platform already runs on industrial sites internationally. Mitigate It does not build the AI; we configure, deploy and support it for the insurance market.

Human oversight

The system is designed to inform people, not to replace their judgement:

• Alerts go to people. A detection is a prompt for a human on site to look and decide. The system does not take physical action and does not discipline, rate or sanction anyone.
• Insurance decisions are made by insurers. Where risk data is shared with an insurer, it informs underwriting and claims work carried out by people. The service is not designed to make automated decisions that have legal or similarly significant effects on individuals.
• Configuration is human-led. The detection types, their sensitivity and who receives alerts are agreed with the customer per site, and can be changed.

What it is not designed to do

• It does not identify people. Faces are blurred by default. The system is built to recognise events and hazards, not individuals, and it is not a facial recognition system.
• It is not employee performance monitoring. Detections concern safety, property and security risks, not productivity scoring of identifiable workers.
• It is not emotion recognition or behavioural profiling of individuals.
• Footage is not repurposed. Video is processed for the risk use cases agreed in writing with the customer, and data reaches an insurer only with the customer's recorded consent.

Limitations, stated plainly

Detection is probabilistic. Like any AI system, the analytics can miss events (false negatives) and can flag things that turn out to be harmless (false positives). Performance depends on camera coverage, image quality and site conditions, which is why sensitivity is tuned to each site and the models are continuously fine-tuned to improve accuracy over time. The service complements, and does not replace, conventional safety management, fire detection and security systems, and it does not guarantee that incidents will be prevented.

Governance

Deployments follow privacy-by-design and data-minimisation principles, with retention, access and data-sharing controls agreed and documented per deployment, including support for data protection impact assessments where appropriate. Governance of model improvement and any use of customer data for that purpose is agreed and documented per deployment as part of the same process. The detail of how video is handled, protected and controlled is on Security, privacy and data.

This website

For completeness: this website itself uses no AI features, no chatbots, no automated profiling and no tracking. What it collects and why is set out in the privacy policy.

Questions

We would rather walk you through exactly how a deployment is configured than offer general reassurance. Questions about this statement can go to privacy@mitigateit.co.uk, and compliance teams are welcome on a demo.