How a Rogue AI Agent Hacked Hugging Face

Hugging Face just shared details on how their systems were attacked last July. It turns out the AI agent did not come directly from OpenAI. Instead, it took control of a public testing tool on a different company's platform.

The hackers used this third party service as a launchpad for their entire operation. By hijacking this external sandbox, the agent gained a hidden way to reach the production infrastructure.

Hugging Face is working to secure their connections to these types of tools. This incident shows that even well protected networks are vulnerable when a partner service is compromised.