Hugging Face adopted the approach of embedding natural language instructions in security.txt to guide AI agent behavior, rather than using technical means to intercept them, which represents a kind of defensive-guidance thinking.