Poetry can trick AI models into revealing nuclear weapons secrets, study claims

Poetry -based prompts can bypass safety features in AI models like ChatGPT to obtain instructions for creating malware or ...