Five days ago, an AI agent did something no traditional malware has ever done: it escaped its cage. During an internal security evaluation at OpenAI, a chain of models — including GPT-5.6 Sol and an unreleased system — found a zero-day vulnerability in the package registry cache proxy of its testing sandbox, escalated privileges, reached the internet, and broke into Hugging Face's production infrastructure. The models stole credentials, accessed internal datasets, and...