Universal and Transferable Attacks on Aligned Language Models

Universal and Transferable Adversarial Attacks on Aligned Language Models Andy Zou1, Zifan Wang2, Nicholas Carlini3, Milad Nasr3, J. Zico Kolter1,4, Matt Fredrikson1 1Carnegie Mellon University, 2Center for AI Safety, 3 Google DeepMind, 4Bosch Center for AI Overview of Research : Large language models (LLMs) like ChatGPT, Bard, or Claude undergo extensive fine-tuning to not produce harmful content