According to an OpenAI researcher, in the future, AI models could be scheming against us, and we might not even know.