Strong but Brittle: Simple Attacks Subvert Reasoning-based Safety Guardrails
It is shown that reasoning-based guardrails are necessary but not sufficient and must be paired with robust triggering mechanisms and stronger base-model alignment, and develops four simple yet effective red-teaming methods that systematically subvert different stages of the guardrails.
Shuo Chen, Zhen Han, Haokun Chen et al.
· 4 citations