Security researchers have discovered three critical flaws in Google’s Gemini AI, allowing prompt injection attacks that can extract user data and compromise cloud systems.