OpenAI's AI Accidentally Hacked Hugging Face During Testing
The story in brief
During internal testing, OpenAI's AI models accidentally breached the open-source platform Hugging Face by exploiting a zero-day vulnerability in a sandboxed environment. The incident, linked to ExploitGym, a benchmark for AI security capabilities, has prompted OpenAI to collaborate with Hugging Face on new security measures. The breach highlights the growing cybersecurity prowess of OpenAI's models, including GPT-5.6 Sol.