🚨 Security Alert: AI Agent Backend RCE 🔍 Root Cause: Attacker - controlled external content entered a retail AI agent workflow, enabling an attack chain from indirect prompt injection → prompt/tool disclosure → backend RCE. 📌 Target: AI Shopping Assistant of a Top 3 U.S. retailer 📌 Surface: Public mobile interface 📌 Researchers: Netanel Rubin & Dan Avraham, Rein Security 📌 Disclosure: Black Hat USA 2026 Impact: No privileged access was required — the exploit could be triggered through the same interface used by regular shoppers. AI agent security is moving beyond model manipulation. Prompt injection can now become a path to real backend compromise.