OpenAI’s latest model launch and the cancellation of an updated version over safety concerns have intensified debate over whether AI capabilities could outpace human control