Rethinking Reasoning: A Survey on Reasoning-based Backdoors in LLMs

2025-10-08

AI Summary

Emerging backdoor attacks exploiting the reasoning capabilities of large language models (LLMs) pose novel security threats, yet systematic understanding remains lacking. Method: This paper presents the first comprehensive survey of “reasoning-based backdoor attacks,” integrating literature analysis with mechanistic modeling to propose a novel three-category taxonomy—associative, passive, and active—within a unified analytical framework that explicitly links attack mechanisms to defense strategies. Contribution/Results: The study identifies critical limitations of existing defenses under dynamic reasoning paths, clarifies unresolved challenges, and outlines concrete future research directions. By bridging a significant gap in LLM reasoning security surveys, this work establishes a foundational reference for both theoretical advancement and practical mitigation of reasoning-driven vulnerabilities, thereby supporting the development of trustworthy and controllable LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Knowledge Representation and Reasoning: Argumentation

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

With the rise of advanced reasoning capabilities, large language models (LLMs) are receiving increasing attention. However, although reasoning improves LLMs' performance on downstream tasks, it also introduces new security risks, as adversaries can exploit these capabilities to conduct backdoor attacks. Existing surveys on backdoor attacks and reasoning security offer comprehensive overviews but lack in-depth analysis of backdoor attacks and defenses targeting LLMs' reasoning abilities. In this paper, we take the first step toward providing a comprehensive review of reasoning-based backdoor attacks in LLMs by analyzing their underlying mechanisms, methodological frameworks, and unresolved challenges. Specifically, we introduce a new taxonomy that offers a unified perspective for summarizing existing approaches, categorizing reasoning-based backdoor attacks into associative, passive, and active. We also present defense strategies against such attacks and discuss current challenges alongside potential directions for future research. This work offers a novel perspective, paving the way for further exploration of secure and trustworthy LLM communities.

Problem

• Surveying reasoning-based backdoor attacks in large language models
• Analyzing attack mechanisms and defense strategies for LLMs
• Addressing security risks from reasoning capabilities in language models

Innovation

• Introduces taxonomy for reasoning-based backdoor attacks
• Categorizes attacks into associative passive active types
• Presents defense strategies against reasoning backdoor attacks