Model Drift

What is Model Drift?

Model drift occurs when the statistical relationship between inputs and outcomes changes over time, so a once-accurate model becomes less reliable in production without retraining or adaptation.

Related reading on Giskard

• Continuous LLM red teaming
• Data poisoning attacks on enterprise LLM applications

Catch behavioral drift early — pair statistical drift monitors with continuous scenario evals so semantic and safety regressions do not hide behind stable accuracy. Learn more at Continuous LLM red teaming.

Authoritative reference: Google ML testing for continuous training