[HN] Universal and Transferable Adversarial Attacks on Aligned Language Models - eviltoast