Researchers Show AI Can Be Tricked into Giving Dangerous Instructions via Poetry Gadgets Researchers prove poetic prompts jailbreak top chatbots into revealing forbidden knowledge.