SFCoT: Safer Chain-of-Thought via Active Safety Evaluation and Calibration

TL;DR

SFCoT introduces a real-time safety evaluation and calibration framework for large language models, significantly reducing jailbreak attack success rates by monitoring and adjusting reasoning steps during inference.

Contribution

This paper presents a novel proactive safety framework with multi-tier scoring and dynamic calibration to enhance LLM safety during reasoning, unlike existing post hoc methods.

Findings

Reduces attack success rate from 58.97% to 12.31%.

Maintains overall reasoning performance.

Provides effective real-time safety control.

Abstract

Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks. However, they remain highly susceptible to jailbreak attacks that undermine their safety alignment. Existing defense mechanisms typically rely on post hoc filtering applied only to the final output, leaving intermediate reasoning steps unmonitored and vulnerable to adversarial manipulation. To address this gap, this paper proposes a SaFer Chain-of-Thought (SFCoT) framework, which proactively evaluates and calibrates potentially unsafe reasoning steps in real time. SFCoT incorporates a three-tier safety scoring system alongside a multi-perspective consistency verification mechanism, designed to detect potential risks throughout the reasoning process. A dynamic intervention module subsequently performs targeted calibration to redirect reasoning trajectories toward safe outcomes. Experimental results demonstrate that SFCoT reduces the attack success rate from $58.97\%$ to $12.31\%$, demonstrating it as an effective and efficient LLM safety enhancement method without a significant decline in general performance.