What is Model Drift?
Model drift is the degradation or change in an AI model’s performance over time because real-world data, user behavior, or the operating environment no longer matches the conditions it was trained or validated on. It is significant in compliance because unresolved drift can create accuracy, safety, fairness, and security issues that undermine approved use cases and required controls.
In Depth
In practice, drift appears when input data distributions change, target relationships shift, labels become stale, or the business process around the model changes, causing predictions or outputs to become less reliable. Compliance teams need monitoring, thresholds, incident escalation, retraining governance, and documented change control so they can detect when a model is no longer performing within approved bounds.
It matters especially for regulated deployments such as employment, lending, healthcare, and high-risk decision systems, where poor performance can translate into legal, consumer, or safety harm. Drift monitoring and post-deployment performance review are referenced or implied in frameworks such as ISO/IEC 42001, NIST AI RMF, SOC 2 + AI, the EU AI Act’s lifecycle and monitoring expectations for high-risk systems, and DORA where ICT and model-related changes can affect resilience and incident handling.
Related Frameworks
Related Topics
Related Terms
Weekly digest
Leave your email to get each issue in your inbox. Free, no account required.
We use your email only for the digest. Privacy policy