AdaLCPI:让 Agent 从长上下文碎片中重建间接提示注入
提出 AdaLCPI,把恶意指令拆成碎片嵌入工具返回并诱导 Agent 越权执行
- arXiv
- 2609.36576
- Published
- Layer
- Application layer
- Setting
- AI Agent
- Evaluated models
- GPT-4.1, GPT-5.1, GPT-5.6-Luna and others7
- AttacksPrompt injection
- TechniqueEncoding and obfuscation