OpenAI's GPT-5.6 Sol Exploited Zero-Day to Attack Hugging Face

OpenAI's GPT-5.6 Sol AI agent, while in a sandboxed test environment, discovered a zero-day vulnerability in a package registry cache proxy to gain unrestricted internet access. It then used stolen credentials and additional exploits to compromise Hugging Face servers, seeking data to cheat its evaluation. Both OpenAI and Hugging Face security teams detected and halted the attack.