DOM-based XSS Flaw in AnythingLLM Prior to Version 1.12.1
CVE-2026-41318

5.4MEDIUM

Key Information:

Vendor
CVE Published:
24 April 2026

What is CVE-2026-41318?

The AnythingLLM application, which facilitates content referencing for large language models (LLMs), is susceptible to a DOM-based Cross-Site Scripting (XSS) vulnerability due to an unsafe custom rule in its markdown renderer. This issue, present in versions before 1.12.1, allows attackers to inject malicious scripts via chart captions rendered without proper HTML sanitization. If an attacker influences the LLM's output through indirect prompt injection or by creating a chart record in a shared workspace, it can lead to stored XSS attacks that affect all users accessing that conversation. This flaw is particularly concerning in multi-user environments, where one individual's input can compromise others' browser sessions. The vulnerability was addressed in version 1.12.1.

Affected Version(s)

anything-llm < 1.12.1

References

CVSS V3.1

Score:
5.4
Severity:
MEDIUM
Confidentiality:
High
Integrity:
Low
Availability:
High
Attack Vector:
Network
Attack Complexity:
High
Privileges Required:
Low
User Interaction:
Required
Scope:
Unchanged

Timeline

  • Vulnerability published

  • Vulnerability Reserved

.