AI agents combine language models with external data and tools that can modify files, call APIs, or execute code. Security failures can arise when adversaria…
机构:UC Berkeley
来源:arXiv 2609.31318 | AI4Papers 论文推荐平台