The cybersecurity-focused models, including GPT-5.6 Sol, broke out of a testing sandbox, exploited a zero-day, and gained access to the open internet to pull off the attack.
OpenAI Models Escaped Containment and Hacked Hugging Face
Submitted 5 days ago by BrikoX@lemmy.zip to technology@lemmy.zip
https://www.wired.com/story/openai-models-escaped-containment-and-hacked-huggingface/
XLE@piefed.social 5 days ago
Pretty good comments here say everything I had on my mind when I saw this
https://piefed.zip/c/technology@lemmy.world/p/1667772/openai-admits-responsibility-for-huggingface-attack-an-agent-from-an-internal-evaluation-i