AnythingLLM's in-chat markdown renderer prior to version 1.12.1 contains a cross-site scripting vulnerability due to improper neutralization of input in the alt attribute of images. The vulnerability is specifically in the Chartable component, which renders chart captions derived from LLM-generated natural language without applying DOMPurify sanitization. This allows an attacker capable of influencing LLM output or creating chart records in a shared workspace to inject stored XSS payloads that execute in other users' browsers when they load the chat history. The vulnerability is fixed in version 1.12.1.

Successful exploitation allows an attacker to execute arbitrary scripts in the browsers of other users viewing the affected chat conversations. This can lead to theft of sensitive information, session hijacking, or other client-side impacts. The vulnerability requires the attacker to influence LLM output or create chart records in a multi-user workspace, and user interaction is required to trigger the XSS. The CVSS score of 5.4 reflects a medium severity with high confidentiality impact but lower integrity and availability impacts.

Upgrade AnythingLLM to version 1.12.1 or later, which contains a patch that properly sanitizes chart captions to prevent this XSS vulnerability. Until upgraded, avoid allowing untrusted users to create chart records or influence LLM output in shared workspaces. There is no official vendor advisory provided here, so check Mintplex-Labs' resources for confirmation and additional guidance.