Show HN: AI Security Leaderboard – comparing cyber and CBRN safeguards

A leaderboard has been developed to compare the security of AI models, focusing on their ability to withstand cyber and chemical, biological, radiological, and nuclear threats. The leaderboard uses an automated test suite that subjects models to 1500 jailbreak attempts to measure their vulnerability. The tests aim to identify models that can be manipulated into providing detailed responses to harmful questions. A significant difference in robustness has been found among models, with some performing better than others. The most robust models, such as Fable, have been identified through this testing process. The goal of the leaderboard is to highlight the importance of model security as it becomes increasingly relevant.