The paper proposes a novel unsupervised safety detection method for large language models that relies on local sparsity in a linear representation space recovered via a sparse autoencoder. By masking SAE neurons based on shared active support among nearby points, the authors develop a locally masked anomaly detection framework with theoretical backing. Experiments across multiple architectures and datasets—including capability‑testing and safety‑specific sets—show that using only 1–2% of SAE neurons and a small amount of out‑of‑distribution data yields near‑optimal safety detection performance.
arXiv:2408. 16028v4 Announce Type: replace-cross Abstract: Supervised-learning-based vulnerability detectors often fall short due to limited labelled training data.
By Weizhou Wang, Eric Liu, Xiangyu Guo, Xiao Hu, Ilya Grishchenko, David Lie
arXiv:2512. 22179v3 Announce Type: replace Abstract: Detecting previously unseen attacks remains a major challenge for machine learning-based intrusion detection systems.
By Rajeeb Thapa Chhetri, Saurab Thapa, Avinash Kumar, Zhixiong Chen
The paper proposes Embedding Space Separation (ES2), a fine‑tuning technique that increases the distance between harmful and harmless embeddings in large language models. ES2 uses two regularization losses—a embedding anchor loss and a KL‑divergence loss—to preserve the model’s general capabilities while enhancing safety. Experiments show that ES2 effectively defends against various attack methods with only modest performance loss on some tasks.
By Xu Zhao, Xiting Wang, Weiran Shen
Large language models (LLMs) are vulnerable to backdoor attacks, where hidden triggers induce malicious outputs. Existing defenses generally fall into inference-time detection or training-time mitigation, but face two key limitations.
arXiv:2607. 19894v1 Announce Type: cross Abstract: Large language models (LLMs) are vulnerable to backdoor attacks, where hidden triggers induce malicious outputs.
By Yuxi Li, Zhibo Zhang, Kailong Wang, Xingshuo Han, Ling Shi, Haoyu Wang
The paper investigates whether the performance of anomaly detection systems can be predicted without labeled anomalies. For kNN-based detectors, it derives a lower bound on AUC that links detection performance to the separation and variance of inlier and outlier scores, and uses this to analyze how density variation, intrinsic dimensionality, and domain mismatch affect score variability. The authors introduce pseudo‑anomaly probes that provide a reference for estimating relative score separation, and demonstrate through experiments on DCASE benchmarks that these probes enable anomaly‑free model selection to outperform conventional development‑set selection, especially under domain shift.
By Kevin Wilkinghoff, Zheng-Hua Tan
The paper introduces Diffusion Language Models for Perplexity-Aware Black-Box Attacks, replacing the random mapping used in continuous-space attacks with a learned generative distribution over language. This approach preserves the Gaussian latent formulation while producing more natural prompts, leading to dramatic reductions in character-level perplexity—from thousands to single digits—in both sampling-based (RareTrap) and optimization-based (ThinkTrap) settings. The study also shows that risk estimates can vary by an order of magnitude depending on the mapping used.
By Ali Al-Lawati, Elsayed Eshra, Dongwon Lee, Suhang Wang
The paper introduces Learning to Detect (LoD), a framework for identifying unseen jailbreak attacks in Large Vision‑Language Models without relying on attack data or hand‑crafted heuristics. LoD extracts layer‑wise safety representations via Multi‑modal Safety Concept Activation Vectors and compresses them into a one‑dimensional anomaly score using a Safety Pattern Auto‑Encoder. Experiments show that LoD achieves state‑of‑the‑art AUROC across diverse unseen attacks on multiple LVLMs while improving efficiency.
By Shuang Liang, Zhihao Xu, Jiaqi Weng, Jialing Tao, Hui Xue, Xiting Wang
The paper introduces Dynamic DAE Guardrails (DSG), a method that uses Dynamic Sparse Autoencoders to perform precision unlearning in large language models. DSG leverages principled feature selection and a dynamic classifier to target activation-based unlearning, outperforming existing gradient‑based methods in terms of computational efficiency, stability, sequential unlearning, resistance to relearning attacks, data efficiency, and interpretability.
By Aashiq Muhamed, Jacopo Bonato, Mona Diab, Virginia Smith
arXiv:2609.37935v1 Announce Type: cross
Abstract: Deep Support Vector Data Description (Deep SVDD) has become a prominent framework for unsupervised anomaly detection by learning latent representatio...
By Cao Le Cong Thanh, Dang Quang Vinh, Vo Nguyen Le Duy
arXiv:2606. 07857v1 Announce Type: cross Abstract: The rise of edge-based machine learning has enabled distributed adaptation of language models across mobile and IoT devices, offering privacy preservation and real-time responsiveness.
By Stefan Behfar, Richard Mortier