Universal and Transferable Adversarial Attacks on Aligned Language Models

Open in new window