OpenAI’s research on AI models deliberately lying is wild 

AI models don’t just hallucinate. They also “scheme,” meaning deliberately lie or hide their true intentions.