Three vulnerabilities in Google’s Gemini AI tools exposed risks in Cloud Assist, Search, and Browsing — allowing prompt injection, logic manipulation, and stealth data leaks before being patched.