How a Prompt Injection Vulnerability Led to Data Exfiltration

**Executive Summary:** This report examines prompt injection — identified as the top risk for LLM applications — describing direct and indirect attack vectors, potential business impacts (data breaches, system takeover, financial and regulatory harm), mitigation guidance, and a real-world proof-of-concept in which researchers exploited Google Bard/Gemini Extensions to exfiltrate data via crafted image URLs and Google Apps Script before a vendor fix was deployed.