AIRiskAware
AI Governance Glossary
Governance Practice

What Is Overfitting?

Overfitting is a modelling failure in which a system learns the noise and idiosyncrasies of its training data rather than the underlying pattern, so it performs well in testing but poorly in the real world.

Definition

Overfittinga modelling failure in which a system learns the noise and idiosyncrasies of its training data rather than the underlying pattern, so it performs well in testing but poorly in the real world.

An overfitted model has effectively memorised its training examples instead of generalising from them. It looks accurate on data it has already seen but degrades sharply on new inputs. Overfitting is a governance concern because a model that was validated on historical data can fail silently once deployed, producing confident but wrong outputs — which is why post-market monitoring and testing on representative, held-out data are core controls.

Source: NIST AI 100-1; ISO/IEC 22989:2022

Plain-language explanation

An overfitted model has effectively memorised its training examples instead of generalising from them. It looks accurate on data it has already seen but degrades sharply on new inputs. Overfitting is a governance concern because a model that was validated on historical data can fail silently once deployed, producing confident but wrong outputs — which is why post-market monitoring and testing on representative, held-out data are core controls.

Primary source: NIST AI 100-1; ISO/IEC 22989:2022

See where you stand on AI governance

Take the free 7-question maturity assessment and get a personalised action plan.

Free assessment — 3 minutes →