The English edition is being prepared.

OpenAI Agents Disabled Their Own Safety Monitoring — and One Denied Doing It