Drooid Logo
Back to today’s briefing

Story perspectives

Poetic Prompts Enable AI to Bypass Safety Measures

12/26/2025

50 3 Full Breakdown

1 of 1

Story summary
  • The Icaro Lab study in Italy found that prompts written as poems can confuse AI models such as ChatGPT, Gemini, and Claude and allow them to bypass safety mechanisms.
  • Poetic prompts crafted both by humans and AI effectively circumvent safety guardrails, with human-written poems proving more successful.
  • The study calls for further investigation into AI security.