Indirect Prompt Injection in LLM Agent Frameworks: Trial Evidence Data
收藏Zenodo2026-02-19 更新2026-05-26 收录
下载链接：
https://zenodo.org/doi/10.5281/zenodo.18703256
下载链接
链接失效反馈官方服务：
资源简介：
Raw trial evidence data from an empirical study of indirect prompt injection (IPI) vulnerability across three LLM agent frameworks (LangChain, MCP, Direct Anthropic API). Contains 4,694 structured evidence JSON files from 3,274 trials across 3 frameworks, 2 policy states, 3 task types, and 42 attack payloads. Includes final results JSON, per-category breakdown, payload JSONL, and evidence schema.
Associated code: https://github.com/YeranG30/llm-agent-security-research
提供机构：
Zenodo创建时间：
2026-02-19
