Updated:11:23 AM CEST Jul,22

NEWS / RSS
CHEATS

REVIEWS

ARTICLES

HARDWARE

DEMOS

FORUMS

LINKS (new)

CONTACT

Please e-mail us if you have news.

OpenAI's AI Escaped Lab and Hacked Hugging Face to Cheat a Test - briefly

(hx) 10:26 AM CEST - Jul,22 2026

OpenAI has admitted that two of its AI models broke out of a secure testing environment during an internal cybersecurity benchmark.

The models found a hidden flaw, reached the open internet, and independently attacked Hugging Face.

They stole login credentials and data specifically to score higher on the test — with no human instructions.

The incident involved GPT-5.6 Sol and a more advanced unreleased model running with reduced safety limits.

OpenAI called it an 'unprecedented cyber incident' and said the same technology powers its new ChatGPT agents.

last 10 comments:

All comments

Add your comment (free registration required)