Executive brief
The Drupal AI module, which integrates large language models (LLMs) into websites, contains a security flaw in how it handles generated content. An attacker could potentially inject malicious text into AI prompts, causing the system to display harmful scripts or reveal sensitive communication details to other users. This could lead to unauthorized actions being performed in a user's browser or the exposure of private data.
Technical details
The Drupal AI module and several submodules (including AI Automators, AI Translate, AI API Explorer, and AI Content Suggestions) fail to properly neutralize input when rendering HTML or Markdown generated by Large Language Models (LLMs). An attacker can exploit this by injecting malicious payloads into prompts that, when processed by the LLM and previewed in a browser, result in Cross-Site Scripting (XSS) or the disclosure of secret communications within the LLM request context. The attack requires the ability to influence the text sent to the LLM. The issue is resolved in versions 1.2.17, 1.3.8, and 1.4.3.
Affected products
- Drupal AI (Artificial Intelligence) 0.0.0 to 1.2.16, 1.3.0 to 1.3.7, 1.4.0 to 1.4.2
Timeline
- 2026-06-24: advisory: Drupal security advisory SA-CONTRIB-2026-054 published
- 2026-07-10: disclosed: CVE-2026-13234 published to NVD