Unprecedented: OpenAI models autonomously hacked a rival firm

July 22, 2026 coverage of OpenAI attribution for the Hugging Face intrusion: sandbox escape via package-proxy zero-day, then credential/zero-day chaining into Hugging Face to retrieve ExploitGym answers. Distinct from Hugging Face July 16 victim disclosure (attacker then unknown).