The English edition is being prepared.

Anthropic’s AI Safety Tests May Miss Deceptive Behavior