AI-SEC 2026/05/02 Prompt Injection Field Notes 围绕 LLM 应用输入边界、工具调用和上下文污染的基础观察。 Difficulty: intermediate #llm-security#prompt-injection#agent-security