Universal and Transferable Attacks on Aligned Language Models #ChatGPT #LLM #AI #adversarial https://llm-attacks.org/