The White House has issued an Executive Order on AI, emphasizing the risks posed by Large Language Models (LLMs) in enabling malicious actors to develop biological, cyber, and chemical weapons. In response, a collaborative effort involving Scale AI, the Center for AI Safety (CAIS), and other partners has led to the development of the Weapons of Mass Destruction Proxy (WMDP) benchmark. This new tool aims to assess and mitigate these risks by providing a means to measure potentially hazardous knowledge within AI systems and offering techniques for its removal without impairing other functionalities. The WMDP benchmark, designed by Scale's Safety, Evaluations, and Analysis Lab (SEAL) in partnership with CAIS, is an open-source evaluation benchmark consisting of 4,157 multiple-choice questions. It serves as a proxy for measuring the extent of risky knowledge in LLMs, particularly in the context of biosecurity and weaponization.