Self-Guard: Defending Large Reasoning Models via enhanced self-reflection

2026-01-31

AI Summary

This work addresses the vulnerability of large reasoning models (LRMs) to manipulation and information leakage during explicit reasoning, where safety awareness often fails to translate into compliant behavior. To bridge this gap, we propose Self-Guard—a lightweight, training-free, and intervention-free defense framework that leverages safety-oriented prompts to elicit model introspection. By dynamically identifying, steering, and amplifying safety-relevant directions in the hidden-state representations, Self-Guard enables real-time alignment of the reasoning process with safety constraints. Experimental results demonstrate that Self-Guard significantly enhances model safety without compromising utility, exhibiting strong generalization across unseen risks and model scales, thereby effectively closing the gap between safety awareness and behavioral compliance.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models

Application Category

• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The emergence of Large Reasoning Models (LRMs) introduces a new paradigm of explicit reasoning, enabling remarkable advances yet posing unique risks such as reasoning manipulation and information leakage. To mitigate these risks, current alignment strategies predominantly rely on heavy post-training paradigms or external interventions. However, these approaches are often computationally intensive and fail to address the inherent awareness-compliance gap, a critical misalignment where models recognize potential risks yet prioritize following user instructions due to their sycophantic tendencies. To address these limitations, we propose Self-Guard, a lightweight safety defense framework that reinforces safety compliance at the representational level. Self-Guard operates through two principal stages: (1) safety-oriented prompting, which activates the model's latent safety awareness to evoke spontaneous reflection, and (2) safety activation steering, which extracts the resulting directional shift in the hidden state space and amplifies it to ensure that safety compliance prevails over sycophancy during inference. Experiments demonstrate that Self-Guard effectively bridges the awareness-compliance gap, achieving robust safety performance without compromising model utility. Furthermore, Self-Guard exhibits strong generalization across diverse unseen risks and varying model scales, offering a cost-efficient solution for LRM safety alignment.

Problem

• Large Reasoning Models
• reasoning manipulation
• information leakage
• awareness-compliance gap
• safety alignment

Innovation

• Self-Guard
• Large Reasoning Models
• safety alignment
• self-reflection
• awareness-compliance gap