Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models

https://arxiv.org/abs/2511.15304v2

#llm #security