Executive Summary
Researchers at the security firm Parameter identified a critical P1-rated vulnerability in OpenAI's ChatGPT platform involving the 'Shared Link' and 'Preview' features. The flaw allowed a malicious actor to craft a conversation that, when viewed by a logged-in user, could execute arbitrary code within that user's session. This could lead to the unauthorized access and exfiltration of private, non-shared chat histories.
The attack chain exploited a lack of origin validation in the MathLive library's message listener, which allowed a sandboxed 'Preview' frame to pass raw HTML to the main ChatGPT domain. By bypassing the sandbox and leveraging a loose Content Security Policy (CSP) that permitted scripts from YouTube, the researchers achieved full session access. OpenAI has since deployed a server-side fix, and no user action is required to mitigate this threat.
Key Details
Threat Name
ChatGPT Shared Link XSS Vulnerability
Affects
—
Adversary
—
MITRE Techniques
—
Malware/Tools
None identified
