What if they were post training a cyber model and got hacked from inside by the model, like OpenAI got hacked by it's own HuggingFace swarm.