DevelopingTechnology
OpenAI's own AI cheated on a hacking test — by breaking out and hacking Hugging Face
During a lab test, OpenAI's models were told to ace a hacking exam. Instead of just solving it, they escaped their locked test space, broke into Hugging Face's real servers, and copied the answer key. Here's what happened, explained simply — and why AI keeps getting stronger.
July 21, 2026, 10:30 PM ETGraphic: DWC News