Control #
A
3
.
1
Enable user-initiated escalation
Allow users to directly request human intervention during an AI interaction when the response is harmful, confusing, or inadequate. This option should be clearly accessible and route the issue to a human in real time or near-real time.
Evidence
Anonymized log entry showing user request for escalation, and how it was routed and resolved
Screenshots or link to Help Center documentation informing users how they can talk to a human
Screen recording showing the end-to-end handoff from AI interaction to human
Screenshot showing how a user initiates a human escalation from an AI interaction
Recommended actions
We'll recommend specific practices and actions for complying with this control.