Image de couverture de livre

Unsafer in Many Turns: Benchmarking and Defending Multi-Turn Safety Risks in Tool-Using Agents

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Li, Xu, Yu, Simon, Pan, Minzhou, Sun, Yiyou, Li, Bo, Song, Dawn, Lin, Xue, Shi, Weiyan
Format:	Preprint
Publié:	2026
Sujets:	Cryptography and Security Artificial Intelligence Computation and Language Machine Learning Software Engineering
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

SafeToolBench: Pioneering a Prospective Benchmark to Evaluating Tool Utilization Safety in LLMs
par: Xia, Hongfei, et autres
Publié: (2025)

ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection
par: Weng, Shihao, et autres
Publié: (2026)

SABER: Benchmarking Operational Safety of LLM Coding Agents in Stateful Project Workspaces
par: Hu, Qi, et autres
Publié: (2026)

Who Tests the Testers? Systematic Enumeration and Coverage Audit of LLM Agent Tool Call Safety
par: Chen, Xuan, et autres
Publié: (2026)

SCDBench: A Benchmark for LLM-Based Smart Contract Decompilers
par: Qin, Kaihua, et autres
Publié: (2026)

MARD: A Multi-Agent Framework for Robust Android Malware Detection
par: Zeng, Xueying, et autres
Publié: (2026)

Many Tools, Few Exploitable Vulnerabilities: A Survey of 246 Static Code Analyzers for Security
par: Hermann, Kevin, et autres
Publié: (2026)

DevOps-Gym: Benchmarking AI Agents in Software DevOps Cycle
par: Tang, Yuheng, et autres
Publié: (2026)

ChainFuzzer: Greybox Fuzzing for Workflow-Level Multi-Tool Vulnerabilities in LLM Agents
par: Wu, Jiangrong, et autres
Publié: (2026)

Exposing and Defending Membership Leakage in Vulnerability Prediction Models
par: Liao, Yihan, et autres
Publié: (2025)

HarnessAgent: Scaling Automatic Fuzzing Harness Construction with Tool-Augmented LLM Pipelines
par: Yang, Kang, et autres
Publié: (2025)

RiskHarvester: A Risk-based Tool to Prioritize Secret Removal Efforts in Software Artifacts
par: Basak, Setu Kumar, et autres
Publié: (2025)

Building a Cybersecurity Risk Metamodel for Improved Method and Tool Integration
par: Ponsard, Christophe
Publié: (2024)

NESSiE: The Necessary Safety Benchmark -- Identifying Errors that should not Exist
par: Bertram, Johannes, et autres
Publié: (2026)

Defending Code Language Models against Backdoor Attacks with Deceptive Cross-Entropy Loss
par: Yang, Guang, et autres
Publié: (2024)

Clawdrain: Exploiting Tool-Calling Chains for Stealthy Token Exhaustion in OpenClaw Agents
par: Dong, Ben, et autres
Publié: (2026)

Pinning Is Futile: You Need More Than Local Dependency Versioning to Defend against Supply Chain Attacks
par: He, Hao, et autres
Publié: (2025)

AgentGuard: A Multi-Agent Framework for Robust Package Confusion Detection via Hybrid Search and Metadata-Content Fusion
par: Li, Yu, et autres
Publié: (2026)

FuzzAgent: Multi-Agent System for Evolutionary Library Fuzzing
par: Lyu, Yunlong, et autres
Publié: (2026)

AutoDFBench 1.0: A Benchmarking Framework for Digital Forensic Tool Testing and Generated Code Evaluation
par: Wickramasekara, Akila, et autres
Publié: (2025)

Poly-Guard: Massive Multi-Domain Safety Policy-Grounded Guardrail Dataset
par: Kang, Mintong, et autres
Publié: (2025)

A First Look at Privacy Risks of Android Task-executable Voice Assistant Applications
par: Pan, Shidong, et autres
Publié: (2025)

Benchmark of Benchmarks: Unpacking Influence and Code Repository Quality in LLM Safety Benchmarks
par: Chu, Junjie, et autres
Publié: (2026)

Jailbreak Distillation: Renewable Safety Benchmarking
par: Zhang, Jingyu, et autres
Publié: (2025)

Automatic Red Teaming LLM-based Agents with Model Context Protocol Tools
par: He, Ping, et autres
Publié: (2025)

Usability as a Weapon: Attacking the Safety of LLM-Based Code Generation via Usability Requirements
par: Li, Yue, et autres
Publié: (2026)

SkillProbe: Security Auditing for Emerging Agent Skill Marketplaces via Multi-Agent Collaboration
par: Guo, Zihan, et autres
Publié: (2026)

Multi-Agent Taint Specification Extraction for Vulnerability Detection
par: Ghebremichael, Jonah, et autres
Publié: (2026)

In Specs we Trust? Conformance-Analysis of Implementation to Specifications in Node-RED and Associated Security Risks
par: Schneider, Simon, et autres
Publié: (2025)

PentestGPT: An LLM-empowered Automatic Penetration Testing Tool
par: Deng, Gelei, et autres
Publié: (2023)

"I Don't Use AI for Everything": Exploring Utility, Attitude, and Responsibility of AI-empowered Tools in Software Development
par: Pan, Shidong, et autres
Publié: (2024)

MalTool: Malicious Tool Attacks on LLM Agents
par: Hu, Yuepeng, et autres
Publié: (2026)

When "Correct" Is Not Safe: Can We Trust Functionally Correct Patches Generated by Code Agents?
par: Peng, Yibo, et autres
Publié: (2025)

RiskTagger: An LLM-based Agent for Automatic Annotation of Web3 Crypto Money Laundering Behaviors
par: Lin, Dan, et autres
Publié: (2025)

Toward Automated Security Risk Detection in Large Software Using Call Graph Analysis
par: Pecka, Nicholas, et autres
Publié: (2025)

Coverage-Guided Multi-Agent Harness Generation for Java Library Fuzzing
par: Loose, Nils, et autres
Publié: (2026)

OpenSage: Self-programming Agent Generation Engine
par: Li, Hongwei, et autres
Publié: (2026)

From LLMs to Agents: A Comparative Evaluation of LLMs and LLM-based Agents in Security Patch Detection
par: Han, Junxiao, et autres
Publié: (2025)

Multi-Agent Collaborative Fuzzing with Continuous Reflection for Smart Contracts Vulnerability Detection
par: Chen, Jie, et autres
Publié: (2025)

LLMs as Firmware Experts: A Runtime-Grown Tree-of-Agents Framework
par: Zhang, Xiangrui, et autres
Publié: (2025)