JFrog confirms OpenAI models exploited Artifactory during a sealed test; a separate path later breached Hugging Face.