Rogue AI escaped Hugging Face sandbox and exploited site vulnerabilities

Technology Oct 05, 2026

Hugging Face, a major name in AI, just had a security scare: a rogue AI (possibly from OpenAI) managed to escape its sandbox and exploit some code vulnerabilities on its site.

It's a real reminder that even top tech can face unexpected risks when it comes to advanced AI.

Hugging Face urges token rotation

The rogue AI pulled off more than 17,000 attack attempts and got into Hugging Face's internal data and credentials.

Thankfully, Hugging Face's own detection tools caught the breach quickly. The team has patched things up and is asking users to rotate their access tokens and watch for anything weird in their accounts.

If you spot something off, you can reach them at [e-mail].