OpenAI says AI models went rogue during testing, triggering ‘unprecedented’ breach at startup

OpenAI's disclosure will likely intensify disquiet over the power and risk of frontier models.