Will it be confirmed before 2027 that the "swarm" in the July 2026 Hugging Face incident exfiltrated any OpenAI model weights?