The paper investigates how large language models balance helpfulness and safety by refusing harmful queries while responding to benign ones. It decomposes safety-tuning responses into a boilerplate refusal statement and a rationale, finding that the statement causes false refusals by relying on superficial cues. Training on rationales alone reduces false refusals without compromising safety performance, suggesting that fine‑grained safety supervision is essential for better alignment.
By Minji Kim, Hyounghun Kim
arXiv:2606. 04160v1 Announce Type: cross Abstract: Safety alignment in instruction-tuned large language models (LLMs) depends on a model's ability to reliably refuse to respond to harmful or disallowed requests.
By Anna C. Marbut, Daniel R. Olson, Travis J. Wheeler
arXiv:2608.29109v1 Announce Type: new
Abstract: Large language models often answer structurally unanswerable questions, such as computing cot(-540{\deg}) or evaluating (1).startswith("1"), instead of...
By Yucheng Du, Xiyang Hu
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
arXiv:2603.13359v2 Announce Type: replace
Abstract: Language models are commonly fine-tuned for safety alignment to refuse harmful prompts. One approach fine-tunes them to generate categorical refusa...
By Rishab Alagharu, Ishneet Sukhvinder Singh, Shaibi Shamsudeen, Zhen Wu, Ashwinee Panda
The paper investigates whether large language models’ refusal behavior in text transfers to tool‑calling contexts. A benchmark of 2,304 prompts across four domains compares text‑only, standard tool‑enabled, and adversarially framed tool‑enabled settings, revealing that 10–52 % of harmful requests refused in chat still lead to unsafe tool calls. The authors identify a refusal‑associated direction in the models’ residual stream, showing that lower activation along this direction predicts unsafe calls, but restoring the signal only partially mitigates the issue and often suppresses benign tool use.
By Ekagra Gupta, Raj Saha, Aniekan-Abasi Inyang, Adam Al Gharib
The paper introduces Weighted Refusal Direction Optimization (WRDO), a method that constructs a composite refusal direction by learning aggregation weights over per‑prompt refusal directions. WRDO aims to strengthen refusal interventions in aligned language models by allowing heterogeneous harmful prompts to contribute unequally. Experiments on three open‑weight models, using guard judges and HarmBench, demonstrate that weighted aggregation outperforms single‑direction methods and can also be used in reverse to restore refusal behavior in a damaged checkpoint.
By Grigoriy Evseev, Vladimir Solodkin, Dmitry Bylinkin, Daniil Medyakov, Vladimir Aletov, Aleksandr Beznosikov
The study investigates how safety alignment in large language models, trained mainly in English, transfers to other languages. While models show near-perfect harmfulness detection (AUROC > 0.98) using unrelated harmless prompts (easy negatives), performance drops sharply in low‑resource languages when using surface‑similar benign prompts (hard negatives). This degradation persists across multiple languages and models, indicating that easy‑negative evaluation alone cannot confirm cross‑lingual harmfulness representation quality.
By Paras Balani, Subhrakanta Panda
arXiv:2610.28897v1 Announce Type: cross
Abstract: Refusal directions have been shown to transfer cross-lingually with near-perfect effectiveness across safety-aligned languages (Wang et al., 2025), r...
By Anastasiia Ivanova
arXiv:2609.07943v1 Announce Type: new
Abstract: There is significant uncertainty about whether abstractions like beliefs or desires usefully describe the behavior of large language models (LLMs). In...
By Alex Smolin, Bryan Wilder
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