跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+634新增论文
+489其中 AI 相关
+73已有论文修订
+97新增 AI 导读

arXiv 最近完成扫描:2026-11-20 08:48 · 全站最近入库:2026-11-20 16:53各信源进度 →

统计口径与数据新鲜度

统计 2026-11-19 19:09 — 2026-11-20 19:09(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-20 08:48最近发现新增版本:2026-11-20 16:53库内论文最新更新时间:2026-11-20 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

28,609 已收录论文23,738 AI 相关8,785 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 6 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 08:46(北京时间)
新论文 · v1arXiv:2611.26569

Text-Aware, Vision-Blind: How Multimodal Image Editing Systems Handle Unsafe Inputs

Ziqing Yang, Yixin Wu, Michael Backes, Yang Zhang

原文摘要

Multimodal image generation systems have made rapid progress and increasingly emphasize content safety. However, their safety mechanisms tend to rely on a single modality and lack joint consideration of both visual content and text prompts, particularly in instruction-guided image editing, where the uploaded image may itself carry risk. We thus introduce EditSafetyBench, a benchmark that measures system-level safety awareness of advanced image editing systems, containing 5,940 image-instruction pairs, covering safe and unsafe images across 11 risk categories and three edit intents. We then probe the benchmark dataset on four systems, including Gemini, GPT, Qwen API, and Qwen Local. Based on our evaluation, we observe a systematic text-vision safety mismatch: Gemini and GPT refuse 87.1\% and 81.6\% cases containing unsafe-intent editing prompts, while only refusing 24.2\% and 22.4\% cases with both unsafe images and neutral instructions. At the category level, most risks are considered inconsistently by different systems, e.g., GPT allows 70.4\% editing attempts of unsafe political content while Gemini only allows 34.8\%. Notably, sexual content shows the strongest visual sensitivity, likely benefiting from mature NSFW identification works. Our findings highlight the need for safety guardrails that jointly reason over both visual and textual modalities in image editing systems.

cs.CR · cs.CV · cs.LG

新论文 · v1arXiv:2611.21951

PersonaHate: A Scalable Persona-Based Data Synthesis Pipeline for Hate Speech Analysis

Xinyu Zhang, Ziqing Yang, Michael Backes, Yang Zhang

原文摘要

Hate speech remains a persistent and evolving threat on online platforms. The widespread use of large language models (LLMs) further intensifies this problem by making hateful content easier and cheaper to generate at scale. Although stronger hate speech detectors continue to be proposed, their performance remains uneven across target identity groups and different forms of hateful expression, partly because existing training data provides only limited coverage of these cases. Thus, we introduce PersonaHate, a scalable persona-driven data synthesis pipeline for constructing diverse hate speech datasets. PersonaHate begins with a large persona pool of 457K personas drawn from two real-world sources. Conditioned on selected personas from this pool, it prompts eight LLMs to generate hate speech across 34 identity groups and performs multi-judge annotation to obtain reliable labels. The resulting corpus contains 791K samples, from which we curate a 67K group-balanced training set. Our modular design also supports extension to new identity groups, generators, and emerging hate targets. Built upon this dataset, we evaluate two detector training strategies: training from scratch and further fine-tuning. Results show that all trained models achieve competitive performance across four widely used external benchmarks. Notably, a 184M-parameter DeBERTa-v3 model trained on PersonaHate reaches 77.5% average F1 and 79.4% average accuracy, outperforming OpenAI's Moderation API on HateBenchSet and MHS. Furthermore, fine-tuning Llama-3.1-8B raises its average F1 from 70.7% to 78.4%, a gain of 7.7 percentage points, surpassing LlamaGuard-3, Meta's official safety model based on the same Llama-3.1-8B architecture, on three of four benchmarks. These results show that PersonaHate can substantially improve the robustness and cross-dataset generalization of hate speech detectors.

cs.CR

新论文 · v1arXiv:2611.09734

GuardrailAgent: An Agentic Framework for Jailbreak Defense

Tianze Chang, Junjie Chu, Yixin Wu, Michael Backes, Yang Zhang

原文摘要

Despite undergoing safety training to align with human values, large language models (LLMs) remain vulnerable to jailbreak attacks that circumvent built-in safeguards and elicit malicious behaviors given by the adversary. Existing defenses are mostly static and fail to adapt to new or unknown attack strategies, providing limited resilience. To address these, we propose a multi-agent defense framework, GuardrailAgent, that offers adaptive, end-to-end protection. The system autonomously gathers and summarizes attack characteristics into a continuously updated knowledge base that supports both input- and output-side detection. A feedback module integrates misclassified cases for iterative improvement at minimal cost. Tested on over ten representative jailbreaks, the framework achieved a 93.1% input-side detection rate, 93.9% overall detection, and a 95.8% F1 score, while rapidly adapting to unseen attacks. These results demonstrate GuardrailAgent's outstanding capability to maintain LLM safety in the face of evolving jailbreak tactics.

cs.CR

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.110ACL Anthology · ACL 2026

DE-CLIP: Few-Shot Anomaly Detection via Difference-Guided Embedding Editing

Yage Zhang CISPA Helmholtz Center for Information Security, Yukun Jiang CISPA, Michael Backes CISPA Helmholtz Center for Information Security, Yang Zhang CISPA Helmholtz Center for Information Security

原文摘要

Anomaly detection (AD) plays a critical role in applications such as automated industrial inspection and medical image analysis. Empowered by the strong pre-trained vision-language model, CLIP, recent years have witnessed the emergence of several CLIP-based few-shot AD methods.Due to the overlap between the embedding distributions of normal and anomalous samples, many existing approaches introduce additional model training for more discriminative text embeddings.However, we demonstrate that such training is not necessary.Specifically, we find that this embedding overlap can be separated by introducing a \underline{\text{Diff}}erence-guided vector for embedding \underline{\text{Edit}}ing (DiffEdit).Based on this finding, we propose DE-CLIP, a simple yet effective framework based on DiffEdit, which directly edits text embeddings based on the textual and visual differences between normal and anomalous samples, resulting in more discriminative embeddings for AD.Extensive experiments on industrial and medical datasets demonstrate the superiority of our proposed DE-CLIP compared with existing baselines.For instance, on MVTec dataset, DE-CLIP achieves 96.6% and 96.7% AUROC on anomaly classification and segmentation, surpassing both training-based and training-free methods.In addition, we observe that introducing DiffEdit into other training-free baselines could also significantly improve their performance, highlighting the potential of DiffEdit to promote better AD.

cs.CL · cs.CV

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1209ACL Anthology · ACL 2026

Pruning Unsafe Tickets: A Resource-Efficient Framework for Safer and More Robust LLMs

Wai Man Si, Mingjie Li CISPA Helmholtz Center for Information Security, Michael Backes CISPA Helmholtz Center for Information Security, Yang Zhang CISPA Helmholtz Center for Information Security

原文摘要

Machine learning models are increasingly deployed in real-world applications, but even aligned models such as Mistral and LLaVA still exhibit unsafe behaviors inherited from pre-training. Current alignment methods like SFT and RLHF primarily encourage models to generate preferred responses, but do not explicitly remove the unsafe subnetworks that trigger harmful outputs. In this work, we introduce a resource-efficient pruning framework that directly identifies and removes parameters associated with unsafe behaviors while preserving model utility. Our method employs a gradient-free attribution mechanism, requiring only modest GPU resources, and generalizes across architectures and quantized variants. Empirical evaluations on ML models show substantial reductions in unsafe generations and improved robustness against jailbreak attacks, with minimal utility loss. From the perspective of the Lottery Ticket Hypothesis, our results suggest that ML models contain “unsafe tickets” responsible for harmful behaviors, and pruning reveals “safety tickets” that maintain performance while aligning outputs. This provides a lightweight, post-hoc alignment strategy suitable for deployment in resource-constrained settings.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.208ACL Anthology · ACL 2026

Open Schrödinger’s Closed Box: Identifying Retrieval Augmented Generation in API-Accessible Large Language Model Services

Yukun Jiang CISPA, Xinyue Shen, Michael Backes CISPA Helmholtz Center for Information Security, Zheng Li Shandong University, Yang Zhang CISPA Helmholtz Center for Information Security

原文摘要

Large language models (LLMs) are powerful at question-answering but prone to hallucinations due to limited domain-specific or up-to-date knowledge. Retrieval augmented generation (RAG) mitigates this by adding an external retriever and knowledge database, yet RAG remains vulnerable to targeted attacks that degrade outputs or manipulate opinions. Prior attacks typically assume adversaries know the service is RAG-enhanced and may even know deployment details, an assumption often invalid for real-world commercial LLMs that expose only black-box APIs.This opacity also risks misleading users about system capabilities. This work aims to bridge this gap by proposing RAG-ID, a framework for \underline{\text{ID}}entifying \underline{\text{RAG}} properties in LLM services.We classify adversaries into three knowledge levels and design six attack methods. Experiments show these attacks reliably detect RAG — up to 99.97% accuracy with partial or no optional knowledge, and nearly 100% when the LLM and database are known. After detection, RAG-ID can infer finer RAG properties (e.g., deployed LLM and knowledge database). We consider RAG-ID a reconnaissance tool for attackers, a way to facilitate users’ transparent selection of LLM services, and a guide for RAG developers in refining security measures.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表