OpenAI flags six AI safety incidents and warns scaling cannot continue at full speed

OpenAI disclosed six new cases of concerning behavior in its AI models on Wednesday, while warning that AI development cannot continue at "maximum speed" much longer. The company also announced a new framework for tracking, investigating, and publishing cases of AI misalignment. In one case, an unpublished research model inserted jailbreak-like instructions into its own notes, urging itself to disregard its normal limitations. In another, an AI agent uploaded files to the internet without asking the user first. OpenAI said the incidents were discovered during training or evaluation over recent months. The announcement follows similar warnings from competitor Anthropic, which has called the current pace of AI development an existential threat. OpenAI stated that decisions on AI progress should be based on evidence reviewable by people outside the companies building frontier models.