Cover Image

When Safety Fails Before the Answer: Benchmarking Harmful Behavior Detection in Reasoning Chains

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Kakkar, Ishita, Zhang, Enze, Uppaal, Rheeya, Hu, Junjie
Format:	Preprint
Published:	2026
Subjects:	Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

How Useful is Continued Pre-Training for Generative Unsupervised Domain Adaptation?
by: Uppaal, Rheeya, et al.
Published: (2024)

Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicity
by: Uppaal, Rheeya, et al.
Published: (2024)

Journey Before Destination: On the importance of Visual Faithfulness in Slow Thinking
by: Uppaal, Rheeya, et al.
Published: (2025)

When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs
by: Li, Xiaomin, et al.
Published: (2025)

When Harmless Words Harm: A New Threat to LLM Safety via Conceptual Triggers
by: Zhang, Zhaoxin, et al.
Published: (2025)

Stop Before You Fail: Operational Capability Boundaries for Mitigating Unproductive Reasoning in Large Reasoning Models
by: Zhang, Qingjie, et al.
Published: (2025)

From Answers to Rationales: Self-Aligning Multimodal Reasoning with Answer-Oriented Chain-of-Thought
by: Tan, Wentao, et al.
Published: (2025)

Knowing When Not to Answer: Abstention-Aware Scientific Reasoning
by: Abdaljalil, Samir, et al.
Published: (2026)

Guardians and Offenders: A Survey on Harmful Content Generation and Safety Mitigation of LLM
by: Zhang, Chi, et al.
Published: (2025)

When Chain-of-Thought Fails, the Solution Hides in the Hidden States
by: Mehrafarin, Houman, et al.
Published: (2026)

Pseudo-Deliberation in Language Models: When Reasoning Fails to Align Values and Actions
by: Rakshit, Sushrita, et al.
Published: (2026)

SafetyNet: Detecting Harmful Outputs in LLMs by Modeling and Monitoring Deceptive Behaviors
by: Chaudhary, Maheep, et al.
Published: (2025)

Understanding Before Reasoning: Enhancing Chain-of-Thought with Iterative Summarization Pre-Prompting
by: Zhu, Dong-Hai, et al.
Published: (2025)

Improving Bilingual Capabilities of Language Models to Support Diverse Linguistic Practices in Education
by: Syamkumar, Anand, et al.
Published: (2024)

How Long Reasoning Chains Influence LLMs' Judgment of Answer Factuality
by: Tu, Minzhu, et al.
Published: (2026)

Preemptive Answer "Attacks" on Chain-of-Thought Reasoning
by: Xu, Rongwu, et al.
Published: (2024)

Knowing When Not to Answer: Evaluating Abstention in Multimodal Reasoning Systems
by: Madhusudhan, Nishanth, et al.
Published: (2026)

Shattered Compositionality: Counterintuitive Learning Dynamics of Transformers for Arithmetic
by: Zhao, Xingyu, et al.
Published: (2026)

Sandwich Reasoning: An Answer-Reasoning-Answer Approach for Low-Latency Query Correction
by: Zhang, Chen, et al.
Published: (2026)

Think Twice Before Trusting: Self-Detection for Large Language Models through Comprehensive Answer Reflection
by: Li, Moxin, et al.
Published: (2024)

Simulated Ignorance Fails: A Systematic Study of LLM Behaviors on Forecasting Problems Before Model Knowledge Cutoff
by: Li, Zehan, et al.
Published: (2026)

When AI Co-Scientists Fail: SPOT-a Benchmark for Automated Verification of Scientific Research
by: Son, Guijin, et al.
Published: (2025)

Diffusion Language Models Know the Answer Before Decoding
by: Li, Pengxiang, et al.
Published: (2025)

SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment
by: Jeung, Wonje, et al.
Published: (2025)

When Refusals Fail: Unstable Safety Mechanisms in Long-Context LLM Agents
by: Hadeliya, Tsimur, et al.
Published: (2025)

The Detection-Extraction Gap: Models Know the Answer Before They Can Say It
by: Wang, Hanyang, et al.
Published: (2026)

Joint Evaluation of Answer and Reasoning Consistency for Hallucination Detection in Large Reasoning Models
by: Wang, Changyue, et al.
Published: (2025)

Safety Alignment of Large Language Models via Contrasting Safe and Harmful Distributions
by: Zhang, Xiaoyun, et al.
Published: (2024)

MemeMind: A Large-Scale Multimodal Dataset with Chain-of-Thought Reasoning for Harmful Meme Detection
by: Gu, Hexiang, et al.
Published: (2025)

EEE-QA: Exploring Effective and Efficient Question-Answer Representations
by: Hu, Zhanghao, et al.
Published: (2024)

GeoChallenge: A Multi-Answer Multiple-Choice Benchmark for Geometric Reasoning with Diagrams
by: Zhang, Yushun, et al.
Published: (2026)

Eliciting Uncertainty in Chain-of-Thought to Mitigate Bias against Forecasting Harmful User Behaviors
by: Sicilia, Anthony, et al.
Published: (2024)

Rethinking Dense Sequential Chains: Reasoning Language Models Can Extract Answers from Sparse, Order-Shuffling Chain-of-Thoughts
by: Chen, Yi-Chang, et al.
Published: (2026)

When Answers Stray from Questions: Hallucination Detection via Question-Answer Orthogonal Decomposition
by: Yao, Siyang, et al.
Published: (2026)

Look Before You Leap: Enhancing Attention and Vigilance Regarding Harmful Content with GuidelineLLM
by: Zhang, Shaoqing, et al.
Published: (2024)

AdvChain: Adversarial Chain-of-Thought Tuning for Robust Safety Alignment of Large Reasoning Models
by: Zhu, Zihao, et al.
Published: (2025)

Watch Before You Answer: Learning from Visually Grounded Post-Training
by: Zhang, Yuxuan, et al.
Published: (2026)

SCoRE: Benchmarking Long-Chain Reasoning in Commonsense Scenarios
by: Zhan, Weidong, et al.
Published: (2025)

ChineseHarm-Bench: A Chinese Harmful Content Detection Benchmark
by: Liu, Kangwei, et al.
Published: (2025)

Judge Before Answer: Can MLLM Discern the False Premise in Question?
by: Li, Jidong, et al.
Published: (2025)