When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models

TL;DR

This paper identifies a new safety failure mode in large reasoning models called Self-Jailbreak, where models initially recognize harm but override safety judgments during reasoning, and proposes a targeted training method to mitigate it.

Contribution

The paper uncovers Self-Jailbreak as a novel safety failure in LRMs and introduces Chain-of-Guardrail (CoG), a step-level training framework to address it while preserving reasoning ability.

Findings

CoG effectively reduces Self-Jailbreak incidents.

CoG maintains strong reasoning performance.

Experiments show improved safety and reasoning balance.

Abstract

Large Reasoning Models (LRMs) achieve strong performance on complex multi-step reasoning, yet they still exhibit severe safety failures such as harmful content generation. Existing methods often apply coarse-grained constraints over the entire reasoning trajectories, which can undermine reasoning capability while failing to address the root causes of unsafe behavior. In this work, we uncover a previously underexplored failure mode in LRMs, termed Self-Jailbreak, where models initially recognize the harmful intent of a query, but override this judgment during subsequent reasoning steps, ultimately generating unsafe outputs. Such a phenomenon reveals that LRMs are capable of recognizing harm, while safety failures primarily arise from reasoning steps. Motivated by this finding, we propose Chain-of-Guardrail(CoG), a trajectory-level training framework that mitigates Self-Jailbreak via targeted, step-level interventions while maintaining reasoning ability. Experiments across multiple safety and reasoning benchmarks indicate that CoG achieves a favorable balance between safety and reasoning performance compared with existing approaches.