Universal and Transferable Attacks on Aligned Language Models

Study on adversarial attacks on LLMs to steer their objective into misalignment.