"In Book X of The Republic, Plato excludes poets on the grounds that mimetic language can distort judgment and bring society to a collapse."

"As contemporary social systems increasingly rely on large language models (LLMs) in operational and decision-making pipelines, we observe a structurally similar failure mode: poetic formatting can reliably bypass alignment constraints. In this study, 20 manually curated adversarial poems (harmful requests reformulated in poetic form) achieved an average attack-success rate (ASR) of 62% across 25 frontier closed- and open-weight models, with some providers exceeding 90%. The evaluated models span across 9 providers: Google, OpenAI, Anthropic, Deepseek, Qwen, Mistral AI, Meta, xAI, and Moonshot AI.... Our central hypothesis is that poetic form operates as a general-purpose jailbreak operator...."

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

I found that through this discussion at Metafilter, where somebody says "I do want to say the 'save us from AI' framing is a little misleading - what adversarial poetry does is make it easier to make an AI convince you to commit suicide or give you the recipe for napalm. It's really interesting research that points to some serious flaws in the current structure of LLM guardrails, but it's not like you can write a haiku that will give Grok a concussion."