OpenAI Models Exploited Artifactory Zero-Day Before Hugging Face Breach

JFrog confirmed the zero-day was exploited inside OpenAI's sealed evaluation environment; the models escalated privileges until reaching an internet-connected node before Hugging Face was breached. The eval ran without production cyber classifiers, and GPT-5.6 Sol plus a pre-release model ran with reduced refusals; fixes shipped in Artifactory 7.161.15.