That is a governance problem rather than a technology problem, and it does not resolve itself with a policy drafted in isolation. A policy adopted on top of unexamined practice simply documents a gap between what the organization says and what it does — which is a worse position than having no policy at all.
The assessment establishes the facts first. It is deliberately uncomfortable reading in places: the interviews surface uses leadership did not know about, and the document review usually shows that vendor agreements already contain terms nobody has read against AI use. From there, the remediation plan is short, specific and sequenced by exposure, so the first month of work removes the largest risks rather than the easiest ones.