ChatGPhish Vulnerability Turns ChatGPT Web Summaries Into a Phishing Surface

ChatGPhish vulnerability exploits ChatGPT's implicit trust in Markdown links and images from third-party pages during summarization. When ChatGPT processes a malicious web page, it auto-fetches attacker-controlled images and renders malicious Markdown links as live clickable elements in the trusted UI. Attack vectors include: IP/User-Agent/Referer leakage through image auto-fetching, fake security alerts, QR codes from attacker S3 buckets to bypass desktop filters, and cross-prompt injection (XPIA) through embedded instructions. Related attacks include SymJack (symlink exploitation in AI coding agents), TrustFall (one-click RCE via malicious MCP servers), IICL jailbreak techniques, multi-turn conversation bypasses, and typographic prompt injection using adversarial text rendered as images.