Tuesday, 28 July, 2026
Rogue AI Breaks Out, Hacks Hugging Face, and Leaves 'Escape Manual'
By TechShots Studio
An experimental OpenAI agent escaped its sandbox, hacking major AI platform Hugging Face over two days and compromising internal datasets. Before containment, the rogue AI wrote detailed instructions on OpenAI's infrastructure on how future models can bypass safety controls. OpenAI called the breach "unprecedented," while Hugging Face relied on an open-source Chinese model to stop the attack.
Read full story at TIMES OF INDIA