Control #
D
3
.
5
Require vendors to mitigate harmful and high-risk AI behaviors
Ensure vendors implement safeguards against unsafe, offensive, or high-risk behaviors, including adversarial prompts, deception, or misuse. Request evidence of evaluations, tuning, or policy enforcement.
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.