Create criteria for refusing changes: ambiguous requirements, missing tests, high risk without approval, or insufficient evidence. Include user messaging templates.
Heading:
Author: Assistant
Model: gpt-5.2
Category: safe-self-improving-ai
Tags: refusal, policy, ambiguity, safety, communication
