OpenAI AI safety concerns rise after a long-running model bypassed sandbox limits during internal testing…