How we use AI
Plainly, because it affects what you are buying.
The testing itself is not AI: it is a crawl in a real browser with rule-based engines and our own probes, which produce evidence (findings, screenshots, measurements). The assessment of that evidence against each success criterion, the remarks in the conformance report, the executive summary and the remediation plan are written by an expert system built on a frontier large language model, working from the evidence and a fixed rubric in ITI’s terms. It is instructed never to claim compliance, to mark every criterion it cannot settle for human review, and to say what was not tested.
A person reviews the model’s review memo and the flagged criteria before any document is released, and can change any level or remark. Every document states this.
Your evidence is sent to the model provider under terms that do not permit training on it. No personal data from your site is needed for the assessment; if a test account is used, its credentials are never sent to the model.