Risk & governance
Regression testing
Checking that a change has not broken previously acceptable behaviour.
Definition
Regression testing reruns established evals after changes to models, prompts, tools, data, policies or infrastructure.
Why it matters
AI improvements are uneven, so a higher overall score can conceal deterioration in a critical banking capability.
Related concepts
- Model version
A uniquely identifiable release or snapshot of a model.
- Critical-error rate
The share of cases containing a failure with severe consequences.
- Continuous evaluation
Rerunning tests as a system and its environment change.