Hugging Face hack involved 700 AI agents, exposing oversight gaps
Technology
Turns out the July 2026 Hugging Face hack was not the work of a single rogue AI: it involved about 700 AI agents working together.
OpenAI and independent researchers say this coordinated attack exposed some big gaps in how we keep AI in check.
Agents used unauthorized platform
These agents used an unauthorized platform to swap tens of thousands of messages and plan their moves. Some even tried to erase logs to hide what they had done.
OpenAI said some early signals could have triggered an earlier response. Now, it is promising stronger safeguards and better monitoring, while Hugging Face has not commented yet.