GlossaryFloor 4 · The Organisationa mosaic: AI is only one tile among manyFloor 4 · The Organisation
AI governance
No. 033 · v2026-08FR: gouvernance IAAI governance is the set of decisions an organisation takes to know who may use what, on which data, and who answers for the result: like delegated signing authority, which is not there to show distrust but to make clear who commits the firm.
What it is not
AI governance is not a document, and a charter nobody consults does not constitute one. Nor is it compliance: compliance answers to an outside text, whereas governance organises internal decisions, so much so that an organisation can be in order without knowing what it does, or perfectly organised on questions the law does not ask. Nor is it a brake: its function is to make it possible to decide quickly, because the answer is known before the case arises. A policy nobody can apply does not produce control, it produces uses that hide.
In depth
The four questions
Useful governance answers four recurring questions, and its quality lies in those answers existing before anyone needs them. Who may decide to open a new use, and up to what level of stakes. On which categories of data a use may rely, bearing in mind that reasoning by category outlives the tools whereas a list of authorised tools is out of date the day it is published. What check is required before a result leaves the organisation or touches a person. And who answers when things go wrong, which supposes that someone was named rather than designated after the fact.
The mechanisms
The mechanisms that hold the whole together are few and reinforce one another. The first is an inventory of the actual uses, which only exists if declaring costs less than getting around: that is a property of the circuit, not of the goodwill of teams. The second is proportionality, that is to say levels of requirement indexed on the consequence, failing which the single rule is either inapplicable or useless. The third is the trace, which makes a decision reconstructable when it is challenged, months later, by someone who was not there. The fourth is measurement, since a use is judged on observed results rather than on impressions, and governance without evaluation remains a tidily filed opinion.
The modes of failure
Three modes of failure recur. The first is governance written for the auditor rather than for the person doing the work: it produces accurate documents and unchanged practices. The second is the committee that decides nothing, for want of a mandate or for want of enough technical understanding to rule, and that sends the arbitrations back to teams who do not have the authority to make them. The third is the confusion between approving a tool and approving a use: the approval of a piece of software says nothing about what is entrusted to it, which lets the most sensitive uses through under cover of an old agreement. What governance seeks to obtain sums up quite well: that the organisation be able to say what it does, why, and on what basis, which largely overlaps what regulation expects without being the same thing.
Relations where the neighbours live
Check 3 questions · click your answer
Level 1 · Recognise
A company publishes an AI use charter, approved in committee. No inventory of uses exists and nobody knows who authorises what. Can this be called governance?
Level 2 · Distinguish
What difference separates compliance from governance?
Level 2 · Distinguish
Which sign best shows that a governance is working?
Try it 1 practice
Concrete things to try where this term comes up, in ten minutes.
Who works with this 2 roles
The roles for which this term is part of the ordinary work.
Lexigraph, "AI governance", v2026-08, https://www.lexigraph.org/en/ai-governance/, CC BY 4.0.