Glossary
AI Explainability
The ability to provide reasons, in a form suitable for a given audience, for how an AI system arrived at an output.
Last reviewed 2026-08-22
In plain language
Explainability (often paired with interpretability) is technically and legally contested. Some models are intrinsically more inspectable; others rely on post-hoc explanations that may be incomplete or misleading. Legal duties, where they exist, usually require information that a person can use to understand or challenge a decision—not a full mathematical dump of a model. Explainability needs should be set by context: a clinician, a judge, a regulator and a researcher need different artifacts.
Why it matters for AI governance
High-stakes automated decisions raise due-process and consumer-protection questions. Overclaiming explainability can create a false sense of understanding.
Authoritative sources
- Ethics Guidelines for Trustworthy AI (opens in a new tab) — European Commission