Term Kind Topic What it is
Adversarial Evaluation practice Model Evaluation & Red-Teaming Deliberately attempting to make a model behave badly, because a probabilistic system with no fixed expected output cannot be verified by conventional testing.
Evaluation Gate Coverage Gate Scope Ratio, Release Evaluation Coverage metric Model Evaluation & Red-Teaming The share of a feature's real failure surface that its release evaluation actually exercises, which determines whether a passed gate is evidence of safety or evidence that the suite is stale.
Red Teaming a Model practice Model Evaluation & Red-Teaming Adversarial testing of a model or AI system to find inputs that produce harmful, incorrect or policy-violating outputs before users do.