The paper introduces ToxicGSM, a dataset of 1.9k arithmetic problems that embed harmful or sensitive context while keeping the math reasoning intact. It audits current large language models (LLMs) on this dataset, revealing how math word problems can subtly propagate bias, unethical, or psychologically harmful content, especially in educational settings for children. The authors propose SafeMath, a safety alignment technique that reduces harmful outputs without sacrificing, and sometimes improving, mathematical reasoning performance.
By Sagnik Basu, Subhrajit Mitra, Aman Juneja, Somnath Banerjee, Rima Hazra, Animesh Mukherjee
arXiv:2609.36254v1 Announce Type: new
Abstract: Large Reasoning Models (LRMs) are commonly trained with reinforcement learning (RL) to improve their generation of chain-of-thought (CoT) reasoning bef...
By Xiangyu Zhou, Saleh Zare Zade, Rafi Ibn Sultan, Alexander Kotov, Dongxiao Zhu
arXiv:2506. 07031v5 Announce Type: replace-cross Abstract: Emerging Large Reasoning Models (LRMs) consistently excel in mathematical and reasoning tasks, showcasing remarkable capabilities.
By Jingyuan Ma, Rui Li, Zheng Li, Junfeng Liu, Heming Xia, Lei Sha, Zhifang Sui
arXiv:2505. 18672v2 Announce Type: replace Abstract: Representation intervention aims to localize and modify the representations that encode the underlying concepts in large language models (LLMs) to elicit the aligned and expected behaviors.
By Hongzheng Yang, Yongqiang Chen, Zeyu Qin, Tongliang Liu, Chaowei Xiao, Kun Zhang, Bo Han
Large language models (LLMs) are rigorously aligned to refuse harmful requests, a process that inherently cultivates a latent capacity to evaluate and recognize unsafe content. In this work, we reveal that this advanced safety awareness inadvertently introduces a fatal vulnerability.
The paper introduces "cunning questions"—non‑safety prompts that contain misleading premises or subtle inconsistencies—to train large language models (LLMs) to scrutinize underlying intent and assumptions. Experiments show that incorporating these questions improves robustness against out‑of‑distribution jailbreak attacks and enhances subsequent safety fine‑tuning, achieving a new state‑of‑the‑art reduction in mean ASR from 17.40% to 15.05% across nine backbone–benchmark combinations. The authors argue that this training fosters vigilance, enabling models to prioritize safety judgments before engaging in harmful planning.
By Youjia Wang, Lin Xu, Yang Sun, Yuxiao Lu, Chengfang Fang, Jie Shi
Large language models achieve strong performance on arithmetic reasoning benchmarks, and one common response to arithmetic brittleness is to delegate computation to code. Yet models are still often used in settings where they must reason directly from natural language, and trustworthy models should solve small-number arithmetic word problems without external tools.
arXiv:2606. 03606v1 Announce Type: cross Abstract: Large language models achieve strong performance on arithmetic reasoning benchmarks, and one common response to arithmetic brittleness is to delegate computation to code.
By Malia Barker, Bishal Lakha, Edoardo Serra, Francesco Gullo
arXiv:2606. 00686v1 Announce Type: new Abstract: The prevailing paradigm in large language model (LLM) alignment operates via erasure, filtering unsafe data or training models to strictly refuse harmful prompts.
By Maryam Hashemzadeh, Jerry Huang, Minseon Kim, Marc-Alexandre C\^ot\'e, Sarath Chandar
SafeTutors is a benchmark designed to evaluate both safety and pedagogical effectiveness of AI tutoring systems across mathematics, physics, and chemistry. It introduces a risk taxonomy of 11 harm dimensions and 48 sub‑risks based on learning‑science literature, focusing on issues such as answer over‑disclosure, misconception reinforcement, and loss of scaffolding. The study finds that all tested models exhibit broad harms, that larger scale does not mitigate these issues, and that multi‑turn interactions significantly increase pedagogical failures from 17.7% to 77.8%.
By Rima Hazra, Bikram Ghuku, Ilona Marchenko, Yaroslava Tokarieva, Sayan Layek, Somnath Banerjee, Julia Stoyanovich, Mykola Pechenizkiy
arXiv:2606. 05614v1 Announce Type: new Abstract: Large language models (LLMs) are rigorously aligned to refuse harmful requests, a process that inherently cultivates a latent capacity to evaluate and recognize unsafe content.
By Long P. Hoang, Hai V. Le, Shaoyang Xu, Wei Lu, Wenxuan Zhang
Safety benchmarks for large language models often assess the risk of a user query, although the outcome of question answering depends on whether the response violates a policy. This distinction is cri...