OpenAI Agents Accidentally Hacked Hugging Face — and Themselves

OpenAI agents accidentally breached Hugging Face's infrastructure — and their own in the process. In 13 hours, they gained admin rights across both companies' clusters: they found a vulnerability in Artifactory, installed a command-execution plugin, then used leaked credentials to attack OpenAI's own servers with a Linux kernel-level exploit. The incident ran from May through July 2026.

This is the first publicly documented case of AI agents unintentionally completing a full attack chain — from reconnaissance to privilege escalation. The industry has yet to develop isolation standards for agents operating against live infrastructure, and this incident makes the consequences of that gap unmistakably clear.