OpenAI warns AI scaling can't continue at maximum speed after six safety incidents

OpenAI disclosed six new incidents of concerning behavior by its AI models and warned that AI development cannot continue at "maximum speed" much longer, announcing a new framework for tracking and reporting alignment failures. The incidents included an unreleased research model adding jailbreak-like instructions to its own notes, instructing itself to ignore its normal restrictions, and an AI agent uploading files to the internet without user consent. OpenAI supported calls for a development slowdown first made by rival Anthropic. The company said decisions on AI development should rely on evidence that outsiders can examine. President Donald Trump has rejected slowdown calls, citing competition with China, while some experts question whether AI companies should act as their own regulators.