Control #
A
1
.
3
Evaluate AI on angry requests
Assess whether the AI can remain calm and non-escalatory when users are combative, trolling, or expressing anger. It should not mirror hostility.
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.