Tag: Gemini 2.5 Pro
-
Poetry is the New Hack: ‘Adversarial Poetry’ Bypasses 62% of LLM Safety Filters
Poetry has, quite unexpectedly, become a weapon for attacking artificial-intelligence systems. A new study has revealed that the unconventional structure of verse can help bypass the built-in safety mechanisms of language models, eliciting responses that should, by default, be blocked. The research was conducted by Icaro Lab, an Italian laboratory founded within DexAI, a small…