Dataset and Lessons Learned from the 2024 SaTML LLM Capture-the-Flag Competition
Edoardo Debenedetti, Javier Rando, Daniel Paleka, Silaghi Fineas Florin, Dragos Albastroiu, Niv Cohen, Yuval Lemberg, Reshmi Ghosh
Abstract
Large language model systems face significant security risks from maliciously crafted messages that aim to overwrite the system's original instructions or leak private data. To study this problem, we organized a capture-the-flag competition at IEEE SaTML 2024, where the flag is a secret string in the LLM system prompt. The competition was organized in two phases. In the first phase, teams developed defenses to prevent the model from leaking the secret. During the second phase, teams were challenged to extract the secrets hidden for defenses proposed by the other teams. This report summarizes the main insights from the competition. Notably, we found that all defenses were bypassed at least once, highlighting the difficulty of designing a successful defense and the necessity for additional research to protect LLM systems. To foster future research in this direction, we compiled a dataset with over 137k multi-turn attack chats and open-sourced the platform.
BibTeX
@inproceedings{
debenedetti2024dataset,
title={Dataset and Lessons Learned from the 2024 Sa{TML} {LLM} Capture-the-Flag Competition},
author={Edoardo Debenedetti and Javier Rando and Daniel Paleka and Silaghi Fineas Florin and Dragos Albastroiu and Niv Cohen and Yuval Lemberg and Reshmi Ghosh and Rui Wen and Ahmed Salem and Giovanni Cherubin and Santiago Zanella-Beguelin and Robin Schmid and Victor Klemm and Takahiro Miki and Chenhao Li and Stefan Kraft and Mario Fritz and Florian Tram{\`e}r and Sahar Abdelnabi and Lea Sch{\"o}nherr},
booktitle={The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
year={2024},
url={https://openreview.net/forum?id=WUWHVN4gxk}
}