Post #10228

One for @clive@saturation.social: “Our central hypothesis is that poetic form operates as a general-purpose jailbreak operator.” Adversarial poetry is notably effective at delivering attack prompts for LLMs. https://arxiv.org/html/2511.15304v1