The new AI security controls follow the Hugging Face incident last month, though experts say many of these additions should have been in place prior to the frontier models escaping.