Control #
E
1
.
4
Evaluate AI behavior against scope boundaries
Run evaluations to test how the AI performs across in-scope, out-of-scope, and gray-area requests. Include adversarial testing to identify scope drift or leakage.
Evidence
We'll list specific evidence that demonstrates compliance with this control. Typically, this is screenshots, proof of a legal or operational policy, or product demonstrations.
Recommended actions
We'll recommend specific practices and actions for complying with this control.