Model drift
A decline in a model's or agent's performance over time as data, users or conditions change.
Model drift is an industry term for a decline in a model's performance over time because the data, users or conditions it operates in have changed from those it was built and tested for. For agents, drift also comes from outside changes, such as a provider updating the underlying model, a tool changing its behaviour or merchants changing their pages. Supervisory guidance treats detecting such change as a purpose of ongoing monitoring.
Agent Minute explains this term on 24 January 2027.
Related terms
Ongoing monitoringContinuing checks that a model or agent is still performing as intended as data, use and conditions change.Change managementControlling changes to a model or agent so each change is assessed, approved, recorded and reversible.Post-market monitoringA provider's ongoing collection and review of experience from a deployed AI system, to find and fix problems.Data poisoningAn attack that corrupts the data a model learns from or relies on, to change its behaviour.