Performance Decay from Data Staleness
資料時效落差導致效能衰退
Continuous Validation
Risk Description
When after structural change in the external environment, the model's predictive accuracy declines progressively; due to unmitigated control gaps, lacking a continuous validation mechanism, the organization judges the system normal by its existing metrics and identifies the problem only once cumulative losses become evident, triggering compliance exposure and operational reputational costs.
Framework Mappings
EU AI ActArt.15、Art.72
NIST AI 600-1
ISO/IEC TR 24027
ISO/IEC 5338持續確效
MIT AI Risk RepositoryDomain 7
Risk Treatment & Implementation Guidance
Continuously validate model performance against samples reflecting the current environment rather than historical metrics alone; Monitor input-data distributions and external shifts, triggering revalidation upon structural change; Manage training-data freshness with defined retraining review cycles and triggers