AI Governance Glossary

Enterprise AI programs move faster than policy templates. Boards ask for roadmaps, regulators ask for evidence, and product teams ship models before security has a standard checklist. This glossary exists so you can speak one shared language across legal, risk, procurement, data science, and line-of-business sponsors—without translating every acronym in every meeting.

Each entry starts with what a term means in practice, why it matters to a CAIO or AI governance lead, a concrete example, and where teams commonly misunderstand it. Where we have deeper material, you will also see industry-specific notes for healthcare, finance, manufacturing, government, technology, and insurance, plus how the idea shows up in frameworks such as the NIST AI Risk Management Framework and the EU AI Act. Use the alphabetical browse when you are mapping a policy to controls, or use search when you are responding to a vendor questionnaire and need a precise definition fast.

If you are building an AI inventory, third-party model review, incident response playbook, or board-ready risk summary, consistent definitions reduce rework. They help you separate vendor marketing language from controls you can test—for example, the difference between “explainability” as a demo feature and explainability as an auditable decision trail under model risk management expectations.

Governance is not a single document; it is a chain of decisions: what data is allowed into training, how performance is monitored after release, who approves a new tool, and what happens when outputs drift or harm appears. The terms here connect those decisions to the vocabulary regulators, insurers, customers, and employees already use when they evaluate your program.

Start with the concepts your organization debates most—bias testing, human oversight, DPIA-style impact analysis, logging and traceability, high-risk system obligations—then follow related terms at the bottom of each page to explore adjacent ideas. Over time, this becomes a lightweight reference library your whole program can rely on instead of fifty different slide decks with fifty different definitions.