PostMessage Misconfiguration + AI Prompt Injection + Sandbox Escape = XSS & Data Exfiltration

A detailed writeup of a three-vulnerability exploit chain targeting an AI assistant platform. The chain combines: (1) postMessage misconfiguration using wildcard origins on both sending and receiving sides, (2) AI prompt injection via a URL query parameter that forces HTML rendering in a sandboxed iframe, and (3) a sandbox escape using window.name persistence across cross-origin navigations. Together, these allow an attacker to inject arbitrary scripts into the sandboxed iframe, traverse the opener frame hierarchy, and continuously exfiltrate sensitive data from AI conversations. Full proof-of-concept code is provided along with five concrete fixes including origin validation, DOMPurify sanitization, explicit postMessage target origins, and the Cross-Origin-Opener-Policy header.