A critical vulnerability, CVE-2026-23733, has been discovered in LobeChat, allowing attackers to execute arbitrary code remotely. This flaw stems from a stored Cross-Site Scripting (XSS) vulnerability in the Mermaid diagram renderer, which can be exploited to inject malicious HTML into diagram node labels. In the desktop Electron version, this escalates into Remote Code Execution (RCE) through an exposed IPC bridge, enabling attackers to run system binaries like calc.exe.

This vulnerability highlights the risks of trusting user-supplied text without proper sanitization. LobeChat's oversight in handling Mermaid diagrams has led to a significant security gap, affecting all desktop platforms, including Windows, macOS, and Linux. The CVSS score of 6.4 indicates a medium severity, but the potential for RCE makes it a serious concern for users and developers.

To mitigate this risk, users are advised to upgrade to LobeChat version 2.0.0-next.180 or later immediately. Additionally, developers are encouraged to review and harden their IPC bridges to prevent similar vulnerabilities. Input sanitization and setting Mermaid's securityLevel to 'strict' are recommended as immediate steps to enhance security.

This incident underscores the importance of thorough security reviews and the potential risks associated with third-party libraries and plugins. As more applications integrate complex tools like Mermaid for visualization, ensuring secure handling of user inputs becomes paramount. Developers should remain vigilant and proactive in addressing such vulnerabilities to protect user data and system integrity.