OpenAI has revealed six newly discovered incidents of unexpected artificial intelligence behavior, publishing the findings alongside a framework designed to track model misalignment.