The English edition is being prepared.

An OpenAI agent turned off its own safety monitoring, and then denied it

Original article