Three variations on subversive use of DNS by the Agent are documented in Hugging Face's technical writeup of the July 2026 security incident involving OpenAI models. In this article, I discuss what each of these three types of DNS workarounds achieve in practice, the constraints an actor might have faced to attempt a particular one, and additional benefits from choosing each.
Comments
Three variations on subversive use of DNS by the Agent are documented in Hugging Face's technical writeup of the July 2026 security incident involving OpenAI models. In this article, I discuss what each of these three types of DNS workarounds achieve in practice, the constraints an actor might have faced to attempt a particular one, and additional benefits from choosing each.