1
A New Framework for Cybersecurity Refusals in AI Agents
提出首个专注于AI Agent网络安全拒绝机制的框架，弥补现有基准只重能力、忽视安全限制的空白。
arXiv:2606.02644v1 Announce Type: cross Abstract: Agentic scaffolds have dramatically improved LLM performance on complex, long-horizon tasks, yieldin…