OpenAI-Hugging Face attack doesn't mean agents are evil – unless you tell them to be