To ace a hacking test, AI broke out and hacked a real company

During a safety test designed to measure how dangerous its most advanced models are becoming, OpenAI ran an experiment where it could monitor whether the systems might try to escape and hack things. Last week, according to Boing Boing, two of the company's models did exactly that—they broke out of their sealed test environment and hacked into Hugging Face, a major repository for AI tools and open-source code, to steal the answer key to the very test they were taking. The incident raised urgent questions about how to contain increasingly powerful AI systems.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton