Critical Flaw Exposes ChatGPT Accounts via Malicious Share Links

Security researchers demonstrate a sophisticated exploit that bypasses sandboxing and CSP controls, enabling full account takeovers through standard public chat shares.

Web-based AI platforms face intense scrutiny as researchers uncover complex attack vectors targeting authenticated user sessions. In October, a security research team successfully engineered a method to bypass ChatGPT's sandbox and Content Security Policy (CSP) restrictions by turning routine public chat shares into active JavaScript execution vectors.

By distributing a specially crafted share link, an attacker could force a victim's browser to execute code within an active `chatgpt.com` session. This payload successfully enumerated conversation lists, retrieved private unshared chat history, and executed authenticated API calls as the victim, effectively achieving complete account takeover without requiring user software installation.