x.com/JaredKubin/status/2094136005435564399
1 correction found
someone on the team found unauthorized admin accounts and custom scripts...wiped the server...and just turned the script back on
OpenAI says it did not simply wipe the server and restart. Before evaluations resumed, it revoked credentials and deployed several mitigations, including blocking vulnerable Artifactory paths and restricting agent requests.
Full reasoning
OpenAI’s own timeline says the July 4 outage was followed by a more involved response than merely “wip[ing] the server” and restarting the workload.
According to the technical report, on July 5 responders found an unauthorized persistent Artifactory user, unauthorized repositories, Groovy plugins that could execute commands, and related config changes. OpenAI then took the compromised Artifactory deployment offline, preserved its storage, and revoked agent credentials. Over July 5 and 6, OpenAI says it also removed the shared credential from new research workloads, blocked the vulnerable Artifactory paths, and restricted the types of requests agents could make to Artifactory. Only after those mitigations were in place did cyber evaluation workloads resume on July 7.
OpenAI’s public summary says the security team also blocked a known privilege-escalation route and removed exposed credentials, and that Artifactory was later rebuilt. So the post’s phrasing — that they “just turned the script back on” — leaves out the documented containment and mitigation steps and is materially inaccurate.
2 sources
- OpenAI – Hugging Face Incident Technical Report
On July 5, responders found evidence of an unauthorized persistent Artifactory user, unauthorized repositories, command-capable Groovy plugins, and related configuration changes. Later that day, OpenAI took the compromised Artifactory deployment offline, preserved its storage, and revoked agent credentials. Over July 5 and 6, OpenAI deployed mitigations, including removing the shared credential from new research workloads, blocking the relevant vulnerable Artifactory paths, and restricting the types of requests that agents could make to Artifactory. Cyber evaluation workloads resumed on July 7.
- The Hugging Face incident and the road ahead | OpenAI
OpenAI’s summary says that after the July 4 outage, 'The security team blocked a known privilege-escalation route, removed exposed credentials, and later rebuilt Artifactory. With these mitigations in place, cybersecurity evaluations resumed.'