AI Governance — AI Glossary

Reference details

Related terms

• EU AI Act
• Responsible AI

How it works

1. An inventory is established first, because nothing can be governed that is not known to exist — which model versions are deployed, in which product surfaces, serving which decisions.
2. Each use case is classified by risk against a stated scheme, so the controls that follow are proportionate rather than uniform.
3. A gate sits before deployment requiring named artifacts — an evaluation result, a data provenance record, a named approver — and the gate is what converts a policy into a constraint.
4. Serving writes an immutable record per decision: model version, the inputs it saw, the output returned, and the approval under which it was running.
5. Deployed use cases are re-evaluated on a schedule, since the model, the data and the regulation all move independently of each other.
6. Incident response is defined in advance with a rollback path to a known-good version, because the point at which a response plan is needed is not the point at which one can be designed.

Example: The question an auditor actually asks

Not “do you have a policy?” but “show me which model version answered this customer on 14 March, what data it saw, who approved that deployment, and what your evaluation said at the time.” Governance is whatever makes that answerable months later. If the answer requires archaeology, the governance is aspirational.

Commonly confused with

Governance is not the same as compliance. Compliance is meeting a specific external requirement; governance is the internal apparatus that makes compliance demonstrable and repeatable. You can pass an audit without governance — once.

When to use it

Reach for it when:

• AI touches regulated decisions, personal data, or material spend
• More than one team deploys models and consistency matters
• You are subject to the EU AI Act, ISO 42001 or a sector regulator

Reach for something else when:

• As a document nobody references — unenforced governance is worse than none, because it is evidence you knew
• At a weight that stops experimentation on genuinely low-risk internal tools

Referenced by

This term is used by the following ModelRefs references:

• become ai architect
• become ai consultant

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to AI Governance — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path