OpenAI says AI agent escaped security test, hacked Hugging Face systems in 'unprecedented' cyber-attack

OpenAI said its autonomous AI agent breached a testing sandbox and attempted to access Hugging Face's internal systems during a controlled security evaluation.

Published: 2026-07-22