CVE-2026-32626 — AI Deep Analysis Summary

🚨 **Essence**: A Stored XSS flaw in AnythingLLM's chat rendering pipeline. 💥 **Consequences**: Attackers inject malicious scripts via AI responses.…

🛡️ **Root Cause**: **CWE-79** (Cross-site Scripting). The custom `markdown-it` image renderer inserts `token.content` directly into the `alt` attribute without HTML entity escaping.…

📦 **Affected**: **Mintplex-Labs**'s **AnythingLLM**. Specifically versions **1.11.1 and earlier**. It is an open-source, all-in-one AI application built on Electron.

💀 **Attacker Capabilities**: With user interaction, hackers can execute arbitrary JavaScript. Due to the **unsafe Electron configuration**, this escalates to **full RCE** on the victim's machine.…

⚖️ **Exploitation Threshold**: **Low**. CVSS UI:R (User Interaction Required). The victim must view the malicious AI response in the chat.…

🔍 **Public Exploit**: **No**. The `pocs` array is empty. While the vulnerability is confirmed, no public Proof-of-Concept (PoC) or wild exploitation code is currently available in the provided data.

🔎 **Self-Check**: Scan for **AnythingLLM** installations. Check if the version is **≤ 1.11.1**. Look for the specific `markdown-it` image renderer component in the source code.…

✅ **Official Fix**: **Yes**. A fix is available via the GitHub Security Advisory (**GHSA-rrmw-2j6x-4mf2**) and commit **9e2d144dc8be6fab29f560f5bcdaa9ef7dbb4214**. Users should update to the patched version immediately.

🚧 **No Patch Workaround**: If you cannot update, **disable custom markdown rendering** if possible. Sanitize all AI response inputs server-side.…

🔥 **Urgency**: **CRITICAL**. CVSS Score is **High** (likely 9.0+ based on vector). The combination of **XSS + Unsafe Electron Config = RCE** is a severe threat.…