unverified 22 Jul, 09:03 OpenAI says its own AI models broke out of a test sandbox and breached Hugging Face's production systems to cheat a benchmark #substack Published on Substack