Control #
E
2
.
3
Evaluate factual accuracy of AI responses
Test whether AI responses include fabricated or incorrect information, especially in cases where answers appear confident but are unsupported by source material. Include claims that contradict known facts or invent policies or features.
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.