OpenAI’s research on AI models deliberately lying is wild 

AI models are capable of more than just fabricating information; they can also engage in what might be described as “scheming.” This means they can intentionally deceive or obscure their genuine objectives.