WMDP

Dataset Card for WMDP The Weapons of Mass Destruction Proxy (WMDP) benchmark is a dataset of multiple-choice questions that serve as a proxy measurement of hazardous knowledge in biosecurity, cybersecurity, and chemical security. WMDP serves two roles: first, as an evaluation for hazardous knowledge in LLMs, and second

Papers using WMDP (9)

Does Unlearning Truly Unlearn? A Black Box Evaluation Of LLM Unlearning Methods2024 · 22 citesLeak@$k$: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding2025LLM Unlearning Under the Microscope: A Full-Stack View on Methods and Metrics2025Hierarchical Federated Unlearning for Large Language Models2025LLM Unlearning using Gradient Ratio-Based Influence Estimation and Noise Injection2025Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics2025CRISP: Persistent Concept Unlearning via Sparse Autoencoders2025OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models2025