Skip to content

All product videos

Guided AI exposure assessment

2 minutes 15 seconds · Published · English captions available in the player

See an unknown answer become an evidence question, a conflicting answer flagged, and an exposure map where people decide each finding, using fictional data.

Recorded with hypothetical demonstration data. This film preserves the workflow shown at filming; consult the current feature pages for later changes. It is not legal advice or a compliance certificate.

Full video transcript

Transcript from the published English captions, retaining the narration as recorded.

What should a team do when it does not know the answer to an assessment question? EU AI Fit turns the gap into the next useful question. In this fictional example, Northstar is assessing a supplier's tool that ranks applications for support agent roles. Recruiters see the ranked shortlist before they decide whom to interview. The assessment asks about the purpose, the people affected, the data, the safeguards and foreseeable misuse.

Some answers do not fit together. Here, the tool materially influences a decision about people, but the answers say nobody is affected. The assessment flags the conflict, explains why it matters, suggests the evidence to look for, and names the kind of colleague to ask. That is a suggestion, not an assignment. The supplier says it has tested the product, but the team has not seen results for this use. So rather than guess, the contributor records testing as unknown. The assessment then asks which tests cover this use, what the acceptance criteria were, and where the results can be found.

Before anything is saved, the exposure map shows which areas carry a risk, which need more evidence, and which show no material signal. Missing information is not proof that a system is unsafe. It is not proof that it is safe, either. Each finding then needs a decision. Here, the reviewer records that further evidence is required, and why. The finding stays open. A conflict cannot simply be accepted. The answers must be corrected in a new assessment before approval, and the earlier version stays in the history.

My work then links the team back to the open question, so it is not forgotten. Once the exposure map is approved, the next steps are the regulatory role review and then the classification screening. Each one needs its own reviewed decision. These checks cover selected conflicts, not every possible inconsistency. EU AI Fit helps a team find and record risks. It does not certify that every risk has been found, and it is not legal advice.