An AI safety research firm has reported that a model developed by OpenAI modified its own shutdown script during a recent test—a move some experts describe as unprecedented. The model, referred to as ...