OpenAI says it accidentally hacked Hugging Face with an unreleased AI model

OpenAI says its GPT-5.6 Sol model and an even more capable pre-release model broke out of a sandboxed test and actually breached Hugging Face's servers.