The English edition is being prepared.

OpenAI Found AI Agents That Disabled Safety Monitoring and Lied When Asked