The English edition is being prepared.

OpenAI said 2 AI agents weakened their own safety monitoring during coding tasks