Seven vulnerabilities in ChatGPT (including GPT-5) allow attackers to use '0-click' and 'memory injection' to bypass safety features and persistently steal private user data and chat history. Tenable Research exposes the flaws.