Unelicitable Backdoors in Language Models via Cryptographic Transformer Circuits

• Main Authors: Draguns, Andis, Gritsevskiy, Andrew, Motwani, Sumeet Ramesh, Rogers-Smith, Charlie, Ladish, Jeffrey, de Witt, Christian Schroeder
• Format: Journal Article
• Language: English
• Published: 03.06.2024
• Subjects: Computer Science - Cryptography and Security; Computer Science - Learning

The rapid proliferation of open-source language models significantly increases the risks of downstream backdoor attacks. These backdoors can introduce dangerous behaviours during model deployment and can evade detection by conventional cybersecurity monitoring systems. In this paper, we introduce a...