Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents本次捕获版本报告多个接口存在脆弱性,所测防御保护有限;安全性需要与任务效用共同检查。核对原文署名 ↗