Glossary
Updated
Escalation logic acts as a safety valve. Triggers usually include confidence thresholds (when the AI isn't sure of the next step), impact limits (when an action is too expensive or irreversible), or anomaly detection (when the AI's behavior deviates from expected patterns). It ensures that high-stakes decisions always involve a human-in-the-loop.
Example
A legal-assistant agent summarizes 50 contracts autonomously but triggers escalation logic when it encounters an Indemnity Clause with non-standard language, flagging it for a human lawyer to review.