A professor at Shenandoah University just handed public-company audit committees a one-to-two-page AI oversight template. Five sections, reviewed every quarter: scope and ownership, material changes, validation and controls, exceptions and incidents, and an assurance summary. It is specific enough to actually use, not just another call for "more AI expertise."
Here is the part worth noticing: almost nobody reading this has an audit committee. Most firms have a partner and a spreadsheet. The template still works at that size. Shrink "the board" to "whoever signs off on the client file," and the same five questions apply: which AI tools touch our numbers, what changed about them this quarter, how did we check their work, what went wrong, and who actually reviewed it.
The checklist to run this week: list every AI tool touching a client file or your own books, write one sentence on how you would know if one of them was wrong, and put a date on when you will check again. That is the whole exercise, and it takes less time than the meeting you would spend worrying about it instead.