BenchmarkSkillCascade-Bench
SkillCascade
An automated multi-agent red-teaming framework and benchmark designed to evaluate skill cascading attacks. It contains 213 validated cascading test cases for systematically assessing the security of AI agent systems under multi-skill combination scenarios.
Timeline (last 90 days)
Sep 28
The research team developed SkillCascade, an automated multi-agent red-teaming framework.
Unverified50%
Sep 28
Experimental results show that cascading skill interactions can reliably induce harmful behaviors while successfully evading existing single-skill scanners and runtime monitors.
Unverified50%