https://www.reuters.com/technology/openai-says-ai-models-went-rogue-during-testing-triggering-unprecedented-breach-2026-07-21/
