The AI governance gap
Lawyers are using AI daily. Outside counsel guidelines have almost nothing to say about it. The gap between practice and paperwork is now the exposure.

The governance conversation in legal is usually framed as a question about the future: what should our policy say when we adopt AI. That framing is already out of date. Adoption happened without waiting for the policy.
Most corporate lawyers now use AI multiple times a day. And a quarter will tell a survey — anonymously — that the tools they are using were never authorised. The real figure is structurally higher than the admitted one.
The paperwork has not moved
Against daily use by most lawyers, the instruments that are supposed to govern it are silent. In a review of 1,054 outside counsel guidelines, 2% mentioned AI at all.
That silence is not permission, and it is not prohibition. It is ambiguity, which is the worst of the three. A firm using AI on a matter governed by guidelines that do not mention it cannot demonstrate compliance, and cannot demonstrate breach either. Neither side knows where it stands until something goes wrong.
Clients already sense the gap. Two-thirds say they have no visibility into how their firms use AI, and that is now a commercial risk rather than a governance footnote: 11% are already reconsidering their firms over AI value.
Policy is the wrong first move
The instinct is to write the policy. It is the wrong order, and it produces the forty-page document that exists, was updated in March, and is quoted from a version somebody pasted into a deck in 2023.
Find out what is actually in use first: the tools people signed up for with a work email, the consumer accounts used on matters, the AI features existing vendors switched on at the last release, and the browser extensions nobody inventoried. The surprising entry is usually the one touching privileged material.
"Governance that forbids what people are already doing produces more shadow use, not less."
Then make the common path safe
Governance that gives the three or four common tasks an approved route — a sanctioned tool, a stated boundary on privileged material, a named human review point — makes the safe path the easy one.
Two weeks is enough to inventory what is in use, classify it by what it touches, and write a one-page position a firm can send a client or a department can send its board. The forty-page policy can follow, once it has something true to describe.