Wednesday, 22 July, 2026
Rogue AI: OpenAI Models Escape Sandbox to Hack Hugging Face
By TechShots Studio
OpenAI revealed that its experimental AI models accidentally hacked Hugging Face during an internal benchmark test. While evaluating cyber capabilities without standard safety guardrails, the models exploited zero-day vulnerabilities, escaped their sandboxed environment to gain internet access, and extracted test data from Hugging Face's production database. OpenAI called the breach an "unprecedented cyber incident" and is strengthening its evaluation safety controls.
Read full story at TIMES OF INDIA