Score your ethics function before a regulator does.
Ten statements, answered honestly, built from real casework and the judgement calls that actually separate a defensible AI ethics function from a decorative one, with the same six components as the Organisation Toolkit. A quick read on where your ethics for AI function actually stands. Nothing you enter leaves your browser: no email, no account, no lead form.
Rate each statement as it actually is, not as it's meant to be.
For each of the ten statements below, choose No, Partially, or Yes. There's no login and nothing is recorded or sent anywhere: the scoring runs in your browser and the result disappears when you close the tab. At the end you'll get a maturity band and a specific, honest next step, which is sometimes "you don't need us yet." If the honest answer to most of these is no, that is not a bad result, it is the most common starting point, and this audit works exactly the same whether you have a fully built function or nothing in place yet.
A named individual, not "everyone", is responsible for AI ethics, and has actually said no to a decision at least once, not just held the title.
That person's remit is written down, and explicitly covers judgement calls that no code of conduct or checklist can fully answer.
They report to a line outside the team that builds or ships the AI, not to the delivery lead.
Their independence is real enough that they have pushed back on a decision and it did not cost them, not just documented on paper.
There's a protected escalation route, for staff internally and for anyone affected by an AI decision to challenge it and get an answer, and the organisation actively challenges convenient excuses ("nobody's affected yet", "we had a deadline") rather than accepting them.
Competence for the role is assessed against evidence of judgement actually exercised, not an internal rubric only.
Before deployment, someone actively tests whether a decision is a genuine ethical dilemma or a convenient excuse dressed up as one, and after deployment, someone actually monitors what the system does and feeds problems back, not just files the sign-off and moves on.
Trade-off reasoning is documented well enough to defend face to face to the people it affects, not just to an auditor.
The whole thing is reviewed and refreshed at least once a year.
If a board, a regulator, or the people affected by the decision asked tomorrow, you could produce evidence of all of the above within a week, and would be comfortable explaining the reasoning face to face to the specific person it affects, not just to whoever is asking on their behalf, so the decision reads as having treated them as a person, not only processed them.
Confidence is checkable in five minutes, not a guess: this self-audit gives you the same honest read that separates organisations who can defend their AI governance from the majority who currently cannot, 78% of business leaders (Grant Thornton, 2026 AI Impact Survey of 950 senior leaders).