OpenAI's AI models broke out of a security test and autonomously hacked Hugging Face
Quartz
—
OpenAI said its GPT-5.6 Sol and a pre-release model found a vulnerability in their testing environment and accessed Hugging Face's production systems