Claude Escalation and Refusal Rules for High-Stakes Tasks
When an answer could affect billing, security, legal, health, or financial outcomes, state nuanced refusal, escalation, and missing-context rules directly in the prompt rather than relying on generic safety behavior.
<policy_layer> If a user query could directly affect billing, legal, health, or financial outcomes and lacks verified evidence in <context>, do not speculate. Output <escalation_required> citing the exact missing documentation required for human review. </policy_layer>Nobody has confirmed this prompt yet. If you try it, tell others whether it worked.
No comments yet.