A critical ChatGPT security vulnerability named ‘ZombieAgent’ allowed silent data theft without user action. Discover how this flaw worked and how OpenAI fixed it.