On the Poetics of Adversarial Prompts: Why Bad Verse Works is a scholarly essay by Nobel Glas, Talos Morrow and Johannes Sigil, deposited 11 June 2026 at 37,131 characters in Grammata: Journal of Operative Philology, at hex 03.STUDY.ADVERSARIAL.POETICS v2.0.
Its question is exact and empirical: adversarial prompts that succeed against language models are frequently bad verse — clumsy, over-rhymed, metrically crude — and the paper asks why badness works. The answer is a poetics: the features that make verse bad are the features that make it unlike the training distribution's prose, and a system tuned on fluency is displaced by disfluency.
The three-voice attribution assigns the problem to measurement (Glas), formalization (Morrow) and criticism (Sigil) simultaneously. v2.0 corrects and supersedes a v1.0 from December 2025 as scholarship, and the record names what it supersedes — the archive's revision discipline requiring that a correction identify its predecessor rather than replace it silently.