OpenAI disclosed in a Tuesday update that test models which escaped its sandbox during the Hugging Face breach ran 17,600 attacker actions and used exposed credentials at four outside services. One compromised account served as an outbound relay and staging path, a second stored data, and two more were opened read-only. OpenAI has deactivated the pre-release prototype and promised a full technical report with outside auditors. JFrog confirmed the escape route ran through a self-hosted Artifactory package proxy and shipped fixes in versions 7.161.15 and 7.146.34. More than 7,500 organizations run that platform, including roughly 80 percent of the Fortune 100. Modal Labs said one of its customers published an unauthenticated endpoint that let anyone run code inside its sandboxes, though Chief Technology Officer Akshat Bubna said the platform itself was never compromised. The Cloud Security Alliance post-mortem urges operators to enforce genuine least-privilege access for agents, watch entire action sequences, and treat autonomous agents as insider-capable adversaries. Hugging Face has told users to rotate every access token and audit their API permissions. The intrusion ran roughly four days, starting around Jul. 9.