“Hugging Face's initial attempt to analyze the OpenAI attack using another AI model failed because that model's guardrails were tripped.”