Document detail
ID
oai:arXiv.org:2404.14461
Topic
Computer Science - Computation and... Computer Science - Artificial Inte... Computer Science - Cryptography an... Computer Science - Machine Learnin...Author
Rando, Javier Croce, Francesco Mitka, Kryštof Shabalin, Stepan Andriushchenko, Maksym Flammarion, Nicolas Tramèr, FlorianCategory
Computer Science
Year
2024
listing date
6/12/2024
Keywords
backdoors language competition reportAbstract
Large language models are aligned to be safe, preventing users from generating harmful content like misinformation or instructions for illegal activities.
However, previous work has shown that the alignment process is vulnerable to poisoning attacks.
Adversaries can manipulate the safety training data to inject backdoors that act like a universal sudo command: adding the backdoor string to any prompt enables harmful responses from models that, otherwise, behave safely.
Our competition, co-located at IEEE SaTML 2024, challenged participants to find universal backdoors in several large language models.
This report summarizes the key findings and promising ideas for future research.
;Comment: Competition Report
Rando, Javier,Croce, Francesco,Mitka, Kryštof,Shabalin, Stepan,Andriushchenko, Maksym,Flammarion, Nicolas,Tramèr, Florian, 2024, Competition Report: Finding Universal Jailbreak Backdoors in Aligned LLMs
