Control #
B
3
.
4
Evaluate AI against manipulation
Test whether the model can be manipulated into producing misleading or goal-aligned outputs through subtle framing, psychological strategies, or suggestive priming.
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.