AI safety papers at major conferences
Total 51 papers
- Adversarial Hubness in Multi-Modal RetrievalTingwei Zhang, Fnu Suya, Rishi Jha et al.Conference page
- GraphRAG under FireJiacheng Liang, Yuhui Wang, Changjiang Li et al.Conference page
- Hollow-LLM Attack: Computationally Trivial Weights in Zero-Knowledge Verification of LLM InferenceChen Gong, Beijie Liu, Mengyuan LiConference page
- MetaBreak: Jailbreaking Online LLM Services via Special Token Manipulation
- SoK: Robustness in Large Language Models against Jailbreak AttacksFeiyue Xu, Hongsheng Hu, Chaoxiang He et al.Conference page
- URLcoat: Exploiting Web Search Capability to Jailbreak Large Language ModelsYiheng Sun, Linkang Du, Zhou Su et al.Conference page
- WRATH: Turning Watermark Robustness Against Itself via a Watermark-Agnostic Black-Box Invalidation AttackNan Jiang, Juan Hu, Bangjie Sun et al.Conference page
Prompt injection
4- AttnTrace: Contextual Attribution of Prompt Injection and Knowledge CorruptionYanting Wang, Runpeng Geng, Ying Chen et al.Conference page
- Hijacking Large Audio-Language Models via Context-Agnostic and Imperceptible Auditory Prompt Injection
- PromptLocate: Localizing Prompt Injection AttacksYuqi Jia, Yupei Liu, Zedian Shao et al.Conference page
- When AI Meets the Web: Prompt Injection Risks in Third-Party AI Chatbot PluginsYigitcan Kaya, Anton Landerer, Stijn Pletinckx et al.Conference page
- Demystifying the (In)Security of OAuth-based Account Linking in Connector EcosystemsKaixuan Luo, Xianbo Wang, Pui Ho Adonis Fung et al.Conference page
- Investigating the Impact of Dark Patterns on LLM-Based Web AgentsDevin Ersoy, Brandon Lee, Ananth Shreekumar et al.Conference page
- Towards Automating Data Access Permissions in AI Agents
- WebCloak: Characterizing and Mitigating Threats from LLM-Driven Web Agents as Intelligent ScrapersXinfeng Li, Tianze Qiu, Yingbin Jin et al.Conference page
- AI Wrote My Paper and All I Got Was This False Negative:* Measuring the Efficacy of Commercial AI Text DetectorsSeth Layton, Bernardo B. P. Medeiros, Kevin Butler et al.Conference page
- Breaking Free from Ivory Tower: Evaluating and Enhancing Real-world Chinese Underground Adversarial Jargon DetectionZhifan Jiang, Mingxuan Liu, Yue Qin et al.Conference page
- Evaluating Concept Filtering Defenses against Child Sexual Abuse Material Generation by Text-to-Image ModelsAna-Maria Cretu, Klim Kireev, Amro Abdalla et al.Conference page
- MusicShield: Protection for Musicians in the Era of Generative AI
- SoK: Critical Evaluation of Quantum Machine Learning for Adversarial RobustnessSaeefa Rubaiyet Nowmi, Jesus Lopez, Md Mahmudul Alam Imon et al.Conference page
- SoK: Evaluating Jailbreak Guardrails for Large Language ModelsXunguang Wang, Zhenlan Ji, Wenxuan Wang et al.Conference page
Alignment
1- Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?Yuhang Ma, Jie Wang, Zheng YanConference page
- Ensemble Conformal Predictor (EnCP): A New Conformal Predictor with Robustness Guarantees against Data Poisoning AttacksYuxin Yang, Qiang Li, Runyang Feng et al.Conference page
- Exploiting Leaderboards for Large-Scale Distribution of Poisoned ModelsAnshuman Suri, Harsh Chaudhari, Yuefeng Peng et al.Conference page
- Who Taught the Lie? Responsibility Attribution for Poisoned Knowledge in Retrieval-Augmented GenerationBaolei Zhang, Haoran Xin, Yuxi Chen et al.Conference page
- Your Compiler is Backdooring Your Model: Understanding and Exploiting Compilation Inconsistency Vulnerabilities in Deep Learning CompilersSimin Chen, Jinjun Peng, Yixin He et al.Conference page
Privacy and data breaches
11- ARES: Scalable and Practical Gradient Inversion Attack in Federated Learning through Activation Recovery
- Beyond Indistinguishability: Measuring Extraction Risk in LLM APIs
- LLM Unlearning Should Be Form-IndependentXiaotian Ye, Mengqi Zhang, Shu WuConference page
- LLMThief: Evaluating Configuration Leaking Risks in Commercial LLM App Stores
- On the Detectability of Active Gradient Inversion Attacks in Federated Learning
- Recovering and Rehosting Mobile Local LLM Conversations and Contexts via Memory ForensicsHaichuan Xu, David Oygenblik, Runze Zhang et al.Conference page
- Shared Spotlight Meridian: Distributed Sparse Pseudorandom Functions for Scalable Federated Learning
- The Person Behind the Sound: Demystifying Audio Private Attribute Profiling via Multimodal Large Language ModelsLixu Wang, Kaixiang Yao, Xinfeng Li et al.Conference page
- Toward Efficient Membership Inference Attacks against Federated Large Language Models: A Projection Residual Approach
- VerfCNN, Optimal Complexity zkSNARK for Convolutional Neural Networks
- When Designers Meet GenAI: Understanding the Role of Prompt-to-Design Generators in Privacy Dark PatternsJingzhou Ye, Zhaojie Hu, Yao Li et al.Conference page
Supply chain security
2- On the (In)Security of Loading Machine Learning ModelsGabriele Digregorio, Marco Di Gennaro, Stefano Zanero et al.Conference page
- Parasites in the Toolchain: A Large-Scale Analysis of Attacks on the MCP EcosystemShuli Zhao, Qinsheng Hou, Zihan Zhan et al.Conference page
AI and cyber offense and defense
10- Cottontail: Large Language Model-Driven Concolic Execution for Highly Structured Test Input Generation
- deepSURF: Detecting Memory Safety Vulnerabilities in Rust Through Fuzzing LLM-Augmented Harnesses
- Incalmo: An Autonomous LLM-assisted System for Red Teaming Multi-Host NetworksBrian Singer, Keane Lucas, Lakshmi Adiga et al.Conference page
- Knocking on the Front Door: An LLM-Guided Systematic Analysis of DNS Query Processing Vulnerabilities
- LLMs in the SOC: An Empirical Study of Human-AI Collaboration in Security Operations CentresRonal Singh, Shahroz Tariq, Fatemeh Jalalvand et al.Conference page
- PILOT: Command-line Interface Fuzzing via Path-Guided, Iterative Large Language Model Prompting
- RadKey: An LLM-Guided RF Backscatter System for Through-Wall Keystroke Inference
- SpecAuditor: Generating Audit Specifications for LLM-Driven Bug Detection
- Specializing Language Models for Textual Fuzzing via Reinforcement Learning
- StepStone: LLM-Based GPU Kernel Driver Fuzzing via User-Space Libraries
Red teaming
1- DREAM: Scalable Red Teaming for Text-to-Image Generative Systems via Distribution ModelingBoheng Li, Junjie Wang, Yiming Li et al.Conference page