In a statement from OpenAI's head of safety systems Saachi Jain, the latest version 'didn't quite meet the bar'” It had become more persistent in completing tasks, but OpenAI needed to balance that capability against unauthorized behavior.