A new study has shown that prompts in the form of poems confuse AI models like ChatGPT, Gemini and Claude — to the point where sometimes, security mechanisms don't kick in. Are poets the new hackers?