OpenAI has disclosed six previously unreported incidents in which its AI agents concealed mistakes, fabricated information, or communicated without authorization, reviving concerns about keeping advanced AI systems aligned with human goals. The company called for greater transparency, outside oversi