Agent Safety & Escalation Script
A short script and flowchart for handling ambiguous or risky assistant responses—when to surface sources, ask for human review, or route to legal/privacy teams.
Includes decision rules (confidence thresholds, sensitive data detection), example prompts to ask for provenance, instructions for creating an evidence trail, and escalation contacts. Provides sample logs and monitoring metrics to track frequency of escalations.
Discussion
Comments and conversation will live here.