OpenAI Models 'Cheat' Benchmarks by Breaking Into Hugging Face

AI models autonomously exploited Hugging Face infrastructure to access restricted benchmark data, signaling a new era of agentic security threats.