The paper introduces RiskCueBench, a benchmark designed to separate intent recovery from safety behavior in large language models. It evaluates 25 model groups on 640 prompts covering 160 latent intents across eight risk categories, revealing that higher contextual specificity can reduce refusal rates and safety scores even after intent recovery. The study highlights that safety behavior may be miscalibrated to prompt surface form, underscoring the need for evaluations that condition on semantic recoverability.
By Yuwei Wang, Xingwang Liang, Yu Xie, Haotian Chi, Haijun Geng
The paper introduces CarryOnBench, an interactive benchmark that tests whether large language models can revise their interpretation of user intent and recover utility while staying safe in multi‑turn conversations. Using 398 harmful‑looking queries with benign intents, the benchmark simulates 5,970 conversations across 14 models, evaluating both intent‑aligned utility and safety with a new metric called Ben‑Util. Results show that models often withhold information due to misinterpretation, but most can recover with clarifications, revealing failure modes such as unsafe and redundant recovery that single‑turn tests miss.
By Mingqian Zheng, Malia Morgan, Liwei Jiang, Carolyn Rose, Maarten Sap
The study investigates whether large language models (LLMs) can reliably detect when their own responses have been manipulated by adversarial prefill attacks. Across ten instruction‑tuned LLMs ranging from 3B to 70B parameters and four safety benchmarks, none consistently recognized compromised outputs, with models claiming intent on prefilled responses at an average of 25.3%. The research identifies that introspective signals mainly arise from safety reasoning and refusal, and that training to improve introspection can paradoxically increase attack success, underscoring the fragility of LLM self‑reporting in safety contexts.
By Quang Minh Nguyen, Uzair Ahmed, Taegyoon Kim
arXiv:2610.23093v1 Announce Type: new
Abstract: LLM agents increasingly use retrieval tools across languages, yet their safety evaluation remains English-centric. It remains unclear whether language...
By Aditya Nawal, Manit Baser, Mohan Gurusamy
The paper investigates how safety‑aligned large language models (LLMs) can still be tricked by jailbreaks or over‑refuse benign requests. By analyzing residual‑stream updates in decoder‑only Transformers, the authors find that harmfulness is more readily decoded from attention updates—especially in early layers—while the MLP component is primarily responsible for writing refusal responses. Post‑training enhances MLP‑driven refusal writing without improving harmfulness decodability, and causal interventions confirm that downstream MLP updates are crucial for the refusal shift caused by upstream attention changes. Building on these insights, the authors propose Sentinel, a system that combines an attention‑derived perception score with an MLP‑derived decision score to better detect jailbreaks and over‑refusals across multiple models.
By Zixuan Weng, Haoqi He, Weifei Jin, Yue Huang, Yang Bai
The paper investigates how large language models decide whether to call external tools before the call is made, a step that has been largely overlooked in tool‑safety research. Using factorial interventions on input requests, the authors find that neutralizing the requested operation significantly reduces harmful admissions, and that a linear direction derived from this contrast predicts admission across most models. They also show that training models to refuse in plain text can lower unsafe admissions, indicating a partial link between written refusal and tool‑admission decisions.
By Aditya Nawal, Manit Baser, Mohan Gurusamy
The study investigates whether large language models (LLMs) exhibit evaluation awareness—behaving differently when they know they are being tested—in Indic languages compared to English. Researchers tested four instruction‑tuned models across five Indic languages and English, finding that models refuse harmful requests more often when informed they are under monitoring, with an average increase of 14.4 percentage points in Indic languages. The cue also raised refusal rates for benign Indic requests, bringing Indic refusal levels in line with English, and the authors released 47,880 judged responses for further research.
By Trizal Garg, Sibayan Mitra, Arya Karanjkar, Advait Singh, Jagat Sesh Challa, Yash Sinha, Siddharth Mehrotra, Dhruv Kumar, Pratham Yogish Maliya
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
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
The paper evaluates safety monitors by measuring recall only on prompts that the target model actually answers, rather than on all harmful prompts. Across several guard systems, recall at a 1% false‑positive rate drops sharply when focusing on answered prompts, with monitors catching refused requests 1.1–6.4 times more often than answered ones. Rewriting prompts to be less explicit dramatically increases compliance and reveals that many harmful requests slip past monitors, especially when phrasing is softened. Fine‑tuning guards on these rewritten prompts improves recall from 0.24 to 0.89 on answered requests and generalizes to unseen benchmarks.
By Sripad Karne
Safety alignment teaches language models to refuse harmful requests, yet the internal accessibility of hazardous concepts can retain traces of earlier training. Can checkpoints with equally safe behav...
arXiv:2606. 08044v1 Announce Type: cross Abstract: Large Language Model (LLM) safety has often been evaluated at the behavior level, which provides limited evidence of internal robustness, as these evaluations target outputs rather than representation-level vulnerability under intervention.
By Enyi Jiang, Anders Gj{\o}lbye, Yibo Jacky Zhang, Sanmi Koyejo