Image de couverture de livre

AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Andriushchenko, Maksym, Souly, Alexandra, Dziemian, Mateusz, Duenas, Derek, Lin, Maxwell, Wang, Justin, Hendrycks, Dan, Zou, Andy, Kolter, Zico, Fredrikson, Matt, Winsor, Eric, Wynne, Jerome, Gal, Yarin, Davies, Xander
Format:	Preprint
Publié:	2024
Sujets:	Machine Learning Artificial Intelligence Computation and Language
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Improving Alignment and Robustness with Circuit Breakers
par: Zou, Andy, et autres
Publié: (2024)

Security Challenges in AI Agent Deployment: Insights from a Large Scale Public Competition
par: Zou, Andy, et autres
Publié: (2025)

OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents
par: Kuntz, Thomas, et autres
Publié: (2025)

Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs
par: Davies, Xander, et autres
Publié: (2025)

Evaluating Language Model Reasoning about Confidential Information
par: Sam, Dylan, et autres
Publié: (2025)

Boundary Point Jailbreaking of Black-Box LLMs
par: Davies, Xander, et autres
Publié: (2026)

D-REX: A Benchmark for Detecting Deceptive Reasoning in Large Language Models
par: Krishna, Satyapriya, et autres
Publié: (2025)

Existing Large Language Model Unlearning Evaluations Are Inconclusive
par: Feng, Zhili, et autres
Publié: (2025)

How Vulnerable Are AI Agents to Indirect Prompt Injections? Insights from a Large-Scale Public Competition
par: Dziemian, Mateusz, et autres
Publié: (2026)

Applying Refusal-Vector Ablation to Llama 3.1 70B Agents
par: Lermen, Simon, et autres
Publié: (2024)

RedCodeAgent: Automatic Red-teaming Agent against Diverse Code Agents
par: Guo, Chengquan, et autres
Publié: (2025)

Breaking Agent Backbones: Evaluating the Security of Backbone LLMs in AI Agents
par: Bazinska, Julia, et autres
Publié: (2025)

Adversarial Attacks on Robotic Vision Language Action Models
par: Jones, Eliot Krzysztof, et autres
Publié: (2025)

Agent Skills Enable a New Class of Realistic and Trivially Simple Prompt Injections
par: Schmotz, David, et autres
Publié: (2025)

Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing
par: Lin, Justin W., et autres
Publié: (2025)

Instrumental Choices: Measuring the Propensity of LLM Agents to Pursue Instrumental Behaviors
par: Wiedermann-Möller, Jonas, et autres
Publié: (2026)

Can a Bayesian Oracle Prevent Harm from an Agent?
par: Bengio, Yoshua, et autres
Publié: (2024)

HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents?
par: Jiang, Yukun, et autres
Publié: (2026)

Measuring Harmfulness of Computer-Using Agents
par: Tian, Aaron Xuxiang, et autres
Publié: (2025)

HarmTransform: Transforming Explicit Harmful Queries into Stealthy via Multi-Agent Debate
par: Zhu, Shenzhe
Publié: (2025)

Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks
par: Schmotz, David, et autres
Publié: (2026)

Safety Pretraining: Toward the Next Generation of Safe AI
par: Maini, Pratyush, et autres
Publié: (2025)

Evaluating whether AI models would sabotage AI safety research
par: Kirk, Robert, et autres
Publié: (2026)

UK AISI Alignment Evaluation Case-Study
par: Souly, Alexandra, et autres
Publié: (2026)

Mimetic Initialization of MLPs
par: Trockman, Asher, et autres
Publié: (2026)

AcceleratedLiNGAM: Learning Causal DAGs at the speed of GPUs
par: Akinwande, Victor, et autres
Publié: (2024)

Does Refusal Training in LLMs Generalize to the Past Tense?
par: Andriushchenko, Maksym, et autres
Publié: (2024)

QuantSightBench: Evaluating LLM Quantitative Forecasting with Prediction Intervals
par: Qin, Jeremy, et autres
Publié: (2026)

Human-Guided Harm Recovery for Computer Use Agents
par: Li, Christy, et autres
Publié: (2026)

HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
par: Mazeika, Mantas, et autres
Publié: (2024)

AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents
par: Feng, Yunhao, et autres
Publié: (2026)

Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples
par: Souly, Alexandra, et autres
Publié: (2025)

HARP: Measuring Harm Amplification in Multi-Agent LLM Systems
par: Rahman, Md Hafizur, et autres
Publié: (2026)

Towards Low-Resource Harmful Meme Detection with LMM Agents
par: Huang, Jianzhao, et autres
Publié: (2024)

FUSE-ing Language Models: Zero-Shot Adapter Discovery for Prompt Optimization Across Tokenizers
par: Williams, Joshua Nathaniel, et autres
Publié: (2024)

Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning
par: Huang, Benhao, et autres
Publié: (2026)

Why is SAM Robust to Label Noise?
par: Baek, Christina, et autres
Publié: (2024)

Helpful to a Fault: Measuring Illicit Assistance in Multi-Turn, Multilingual LLM Agents
par: Talokar, Nivya, et autres
Publié: (2026)

MIP against Agent: Malicious Image Patches Hijacking Multimodal OS Agents
par: Aichberger, Lukas, et autres
Publié: (2025)

Training a Generally Curious Agent
par: Tajwar, Fahim, et autres
Publié: (2025)