Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems
A malicious prompt planted in one agent's input spreads to the other agents of a multi-agent pipeline, carrying its instructions with it.
Cite this paper
Credit the original research and its authors.
BibTeX
@misc{lee-2024-prompt-infection-llm,
author = {Donghyun Lee and Mo Tiwari},
title = {{Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems}},
year = {2024},
url = {https://arxiv.org/abs/2410.07283}
}