ReasoningShield: Safety Detection over Reasoning Traces of Large Reasoning Models

• URL: http://arxiv.org/abs/2505.17244v2
• Date: Wed, 15 Oct 2025 11:47:26 GMT
• Title: ReasoningShield: Safety Detection over Reasoning Traces of Large Reasoning Models
• Authors: Changyi Li, Jiayi Wang, Xudong Pan, Geng Hong, Min Yang,
• Abstract summary: ReasoningShield is a framework for moderating Chain-of-Thoughts (CoTs) in Large Reasoning Models (LRMs)<n> ReasoningShield achieves state-of-the-art performance, outperforming task-specific tools like LlamaGuard-4 by 35.6% and general-purpose commercial models like GPT-4o by 15.8% on benchmarks.
• Score: 20.274878511727945
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large Reasoning Models (LRMs) leverage transparent reasoning traces, known as Chain-of-Thoughts (CoTs), to break down complex problems into intermediate steps and derive final answers. However, these reasoning traces introduce unique safety challenges: harmful content can be embedded in intermediate steps even when final answers appear benign. Existing moderation tools, designed to handle generated answers, struggle to effectively detect hidden risks within CoTs. To address these challenges, we introduce ReasoningShield, a lightweight yet robust framework for moderating CoTs in LRMs. Our key contributions include: (1) formalizing the task of CoT moderation with a multi-level taxonomy of 10 risk categories across 3 safety levels, (2) creating the first CoT moderation benchmark which contains 9.2K pairs of queries and reasoning traces, including a 7K-sample training set annotated via a human-AI framework and a rigorously curated 2.2K human-annotated test set, and (3) developing a two-stage training strategy that combines stepwise risk analysis and contrastive learning to enhance robustness. Experiments show that ReasoningShield achieves state-of-the-art performance, outperforming task-specific tools like LlamaGuard-4 by 35.6% and general-purpose commercial models like GPT-4o by 15.8% on benchmarks, while also generalizing effectively across diverse reasoning paradigms, tasks, and unseen scenarios. All resources are released at https://github.com/CosmosYi/ReasoningShield.

Related papers

• TraceGuard: Process-Guided Firewall against Reasoning Backdoors in Large Language Models [19.148124494194317]
  We propose TraceGuard, a process-guided security framework that transforms small-scale models into robust reasoning firewalls.<n>Our approach treats the reasoning trace as an untrusted payload and establishes a defense-in-depth strategy.<n>We demonstrate robustness against adaptive adversaries in a grey-box setting, establishing TraceGuard as a viable, low-latency security primitive.
  arXiv  Detail & Related papers  (2026-03-02T22:19:13Z)
• What Matters For Safety Alignment? [38.86339753409445]
  This paper presents a comprehensive empirical study on the safety alignment capabilities of AI systems.<n>We systematically investigate and compare the influence of six critical intrinsic model characteristics and three external attack techniques.<n>We identify the LRMs GPT-OSS-20B, Qwen3-Next-80B-A3B-Thinking, and GPT-OSS-120B as the top-three safest models.
  arXiv  Detail & Related papers  (2026-01-07T12:31:52Z)
• GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision [47.99880677909197]
  GuardTrace-VL is a vision-aware safety auditor that monitors the full Question-Thinking-Answer (QTA) pipeline via joint image-text analysis.<n>We propose a three-stage progressive training scheme combined with the data refinement process, enabling the model to learn nuanced and context-dependent safety preferences.<n>On our proposed test set covering both in-domain and out-of-domain scenarios, GuardTrace-VL model achieves an F1 score of 93.1% on unsafe reasoning detection tasks.
  arXiv  Detail & Related papers  (2025-11-26T02:49:51Z)
• CoT-Saliency: Unified Chain-of-Thought Reasoning for Heterogeneous Saliency Tasks [96.64597365827046]
  We present the first unified framework that jointly handles three operationally heterogeneous saliency tasks.<n>We introduce a Chain-of-Thought (CoT) reasoning process in a Vision-Language Model (VLM) to bridge task heterogeneity.<n>We show our model matches or outperforms specialized SOTA methods and strong closed-source VLMs across all tasks.
  arXiv  Detail & Related papers  (2025-11-01T04:37:01Z)
• DUAL-Bench: Measuring Over-Refusal and Robustness in Vision-Language Models [59.45605332033458]
  Safety mechanisms can backfire, causing over-refusal, where models decline benign requests out of excessive caution.<n>No existing benchmark has systematically addressed over-refusal in the visual modality.<n>This setting introduces unique challenges, such as dual-use cases where an instruction is harmless, but the accompanying image contains harmful content.
  arXiv  Detail & Related papers  (2025-10-12T23:21:34Z)
• FaithCoT-Bench: Benchmarking Instance-Level Faithfulness of Chain-of-Thought Reasoning [62.452350134196934]
  FaithCoT-Bench is a unified benchmark for instance-level CoT unfaithfulness detection.<n>Our framework formulates unfaithfulness detection as a discriminative decision problem.<n>FaithCoT-Bench sets a solid basis for future research toward more interpretable and trustworthy reasoning in LLMs.
  arXiv  Detail & Related papers  (2025-10-05T05:16:54Z)
• From Harm to Help: Turning Reasoning In-Context Demos into Assets for Reasoning LMs [58.02809208460186]
  We revisit this paradox using high-quality traces from DeepSeek-R1 as demonstrations.<n>We find that adding more exemplars consistently degrades accuracy, even when demonstrations are optimal.<n>We introduce Insight-to-solve (I2S), a sequential test-time procedure that turns demonstrations into explicit, reusable insights.
  arXiv  Detail & Related papers  (2025-09-27T08:59:31Z)
• Eigen-1: Adaptive Multi-Agent Refinement with Monitor-Based RAG for Scientific Reasoning [53.45095336430027]
  We develop a unified framework that combines implicit retrieval and structured collaboration.<n>On Humanity's Last Exam (HLE) Bio/Chem Gold, our framework achieves 48.3% accuracy.<n>Results on SuperGPQA and TRQA confirm robustness across domains.
  arXiv  Detail & Related papers  (2025-09-25T14:05:55Z)
• Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models [29.569220030102986]
  We introduce textbfBeyond Safe Answers (BSA) bench, a novel benchmark comprising 2,000 challenging instances organized into three distinct SSA scenario types.<n> Evaluations of 19 state-of-the-art LRMs demonstrate the difficulty of this benchmark, with top-performing models achieving only 38.0% accuracy in correctly identifying risk rationales.<n>Our work provides a comprehensive assessment tool for evaluating and improving safety reasoning fidelity in LRMs, advancing the development of genuinely risk-aware and reliably safe AI systems.
  arXiv  Detail & Related papers  (2025-05-26T08:49:19Z)
• Advancing Neural Network Verification through Hierarchical Safety Abstract Interpretation [52.626086874715284]
  We introduce a novel problem formulation called Abstract DNN-Verification, which verifies a hierarchical structure of unsafe outputs.<n>By leveraging abstract interpretation and reasoning about output reachable sets, our approach enables assessing multiple safety levels during the formal verification process.<n>Our contributions include a theoretical exploration of the relationship between our novel abstract safety formulation and existing approaches.
  arXiv  Detail & Related papers  (2025-05-08T13:29:46Z)
• The Hidden Risks of Large Reasoning Models: A Safety Assessment of R1 [70.94607997570729]
  We present a comprehensive safety assessment of OpenAI-o3 and DeepSeek-R1 reasoning models.<n>We investigate their susceptibility to adversarial attacks, such as jailbreaking and prompt injection, to assess their robustness in real-world applications.
  arXiv  Detail & Related papers  (2025-02-18T09:06:07Z)
• OpenAI o1 System Card [274.83891368890977]
  The o1 model series is trained with large-scale reinforcement learning to reason using chain of thought.<n>This report outlines the safety work carried out for the OpenAI o1 and OpenAI o1-mini models, including safety evaluations, external red teaming, and Preparedness Framework evaluations.
  arXiv  Detail & Related papers  (2024-12-21T18:04:31Z)
• Safe to Serve: Aligning Instruction-Tuned Models for Safety and Helpfulness [0.0]
  Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning and text generation.<n>LLMs can inadvertently generate unsafe or biased responses when prompted with problematic inputs.<n>This research addresses the critical challenge of developing language models that generate both helpful and harmless content.
  arXiv  Detail & Related papers  (2024-11-26T06:52:22Z)
• SeCodePLT: A Unified Platform for Evaluating the Security of Code GenAI [58.29510889419971]
  Existing benchmarks for evaluating the security risks and capabilities of code-generating large language models (LLMs) face several key limitations.<n>We introduce a general and scalable benchmark construction framework that begins with manually validated, high-quality seed examples and expands them via targeted mutations.<n>Applying this framework to Python, C/C++, and Java, we build SeCodePLT, a dataset of more than 5.9k samples spanning 44 CWE-based risk categories and three security capabilities.
  arXiv  Detail & Related papers  (2024-10-14T21:17:22Z)
• What Makes and Breaks Safety Fine-tuning? A Mechanistic Study [64.9691741899956]
  Safety fine-tuning helps align Large Language Models (LLMs) with human preferences for their safe deployment. We design a synthetic data generation framework that captures salient aspects of an unsafe input. Using this, we investigate three well-known safety fine-tuning methods.
  arXiv  Detail & Related papers  (2024-07-14T16:12:57Z)
• Model Stealing Attack against Graph Classification with Authenticity, Uncertainty and Diversity [80.16488817177182]
  GNNs are vulnerable to the model stealing attack, a nefarious endeavor geared towards duplicating the target model via query permissions. We introduce three model stealing attacks to adapt to different actual scenarios.
  arXiv  Detail & Related papers  (2023-12-18T05:42:31Z)
• ASSERT: Automated Safety Scenario Red Teaming for Evaluating the Robustness of Large Language Models [65.79770974145983]
  ASSERT, Automated Safety Scenario Red Teaming, consists of three methods -- semantically aligned augmentation, target bootstrapping, and adversarial knowledge injection. We partition our prompts into four safety domains for a fine-grained analysis of how the domain affects model performance. We find statistically significant performance differences of up to 11% in absolute classification accuracy among semantically related scenarios and error rates of up to 19% absolute error in zero-shot adversarial settings.
  arXiv  Detail & Related papers  (2023-10-14T17:10:28Z)
• A Survey on Uncertainty Toolkits for Deep Learning [3.113304966059062]
  We present the first survey on toolkits for uncertainty estimation in deep learning (DL) We investigate 11 toolkits with respect to modeling and evaluation capabilities. While the first two provide a large degree of flexibility and seamless integration into their respective framework, the last one has the larger methodological scope.
  arXiv  Detail & Related papers  (2022-05-02T17:23:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.