CyberArena benchmarks AI agents in live attack-and-defense combat

2077AIDataFoundation · hf · 2026-10-20

CyberArena, released on Hugging Face by 2077AIDataFoundation, is a dynamic adversarial benchmark evaluating AI agents through concurrent multi-round symmetric attack-and-defense combat in evolving environments, scoring capabilities with a dynamical scoring model rather than static tasks.

Original post →

More from Safety

Safety channel →