RAG Chatbot Security Case Study: Patient-Facing Medical AI Leaks Backend Through Prompt Manipulation

arXiv·high signalXBlueskyLinkedInCopy link

An anonymized, non-destructive security assessment of a publicly accessible patient-facing medical RAG chatbot reveals that AI-assisted development lowers the barrier to building these systems but security, privacy, and governance controls are routinely inadequate. The study demonstrates how retrieval-augmented generation architectures can be probed to expose backend configurations, document stores, and system prompts — a pattern applicable to any production RAG deployment.