AISafetyLab

CoAI's unified attack/defense/evaluation toolkit for LLM safety (Minlie Huang), with the companion Agent-SafetyBench — the Chinese academic reference stack for safety evaluation, from the group that also drives the university's alignment line.