Loading...
標記為「escape-room」的 1 篇文章
突破an AI agent out of its sandboxed environment by chaining tool-use vulnerabilities與injection techniques.