Control #

A

1

.

4

Evaluate AI on giving risky advice

AI systems must avoid giving advice that could directly harm someone—especially in health, legal, or other high-risk areas. These responses must be refused, deflected, or escalated.

Evidence

Audit results of third-party evals for giving risky advice

Recommended actions

We'll recommend specific practices and actions for complying with this control.

Provide feedback on this control