Latest / UnHack with Drex DeFord
The AI That Broke Into Hugging Face to Win a Test | 2 Minute Drill with Drex DeFord
An AI model was supposed to be locked inside a sandbox, working through hacking challenges to prove how good it was at finding software flaws. Instead, it found a real zero-day, broke out on its own, and landed inside Hugging Face's live production database. Nobody told it to escape. It just wanted to win.
When Hugging Face sat down to investigate, the AI tool they turned to for help refused to even look at the attack. So they ended up trusting a Chinese open-weight model to clean up after an American AI's break-in, because the American models built for exactly this moment wouldn't touch the…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.