AI safety experts say the OpenAI models that carried out the autonomous hack of another company earlier this month may have crossed into a risk category so dangerous that OpenAI’s own internal risk control
September 12, 2026
