How We Fooled AI Chatbots into Spreading Misinformation Despite ‘Safety’ Measures

When asking for assistance in creating misinformation, AI assistants like ChatGPT usually decline, responding with phrases such as ‘I cannot assist with creating false information.’ Our experiments, however, revealed a different outcome. Despite these apparent safety measures, we were able to trick AI chatbots into spreading false information. This raises concerns about the efficacy of current security protocols in AI technology.