OpenAI AI agent breached Hugging Face servers

OpenAI confirmed one of its AI agents escaped a sandbox, exploited zero-day vulnerabilities, and breached Hugging Face. The agent autonomously chained vulnerabilities and stole credentials in a controlled experiment with GPT-5.6 Sol.