"Governed AI" has become decoration. It appears on slide decks between the logo and the roadmap, and it rarely survives contact with a real workflow. Here is what it means in our own operating manual, written plainly enough that you could hold us to it.
Vet
Before a tool touches real work, we check three things: what its data policy actually permits, whether it fits the workflow we have rather than the one we wish we had, and what it costs at the volume we would really use. Most of what we test does not make it through.
The discipline is not in the checking. It is in being willing to say "we haven't tested this" out loud, to a client, when saying it is inconvenient.
Disclose
Wherever silence would mislead someone who trusts us, we say what happened. That covers AI-assisted work, partner relationships that pay us, and any published fact that came from a model rather than a source.
Disclosure has a reputation as a compliance tax. In practice it is the cheapest trust-building mechanism available, because it is the one thing competitors imitating your language cannot fake.
Verify
The model drafts. A qualified person decides. Every fact, citation and number gets checked before it reaches a client or goes public — and in finance work, "qualified" means exactly what it says.
This is the step that costs real time, and the one most often quietly skipped. It is also the only reason the first two steps mean anything.
The part most teams miss
None of this slows the work down in the way people expect. It changes where the time goes: less spent producing volume, more spent on the judgement calls that determine whether the volume was worth producing.
The teams getting durable value from AI are not the ones moving fastest. They are the ones whose outputs survive scrutiny — from an auditor, a regulator, a board, or a client who asks a hard question six months later. Speed without that is a liability with good margins, for a while.
Why we publish this
We hold this standard across everything we run, and we would rather be judged against a written version of it than a vague claim to be responsible. If it is useful to you, take it and adapt it. If you think it is wrong somewhere, we would genuinely like to know.
If governed AI adoption is on your plate and the governance part is the bit keeping you up, that is a conversation we have often — [email protected].