Siyuan

A critical missing authorization vulnerability, CVE-2026-66012, in SiYuan before version 3.7.2 allows a remote unauthenticated attacker to exploit the POST /mcp kernel endpoint when the Publish server is in anonymous mode, leading to arbitrary file writes, sensitive credential exposure, malicious plugin execution, and ultimately administrator takeover on affected systems.

A critical vulnerability (CVE-2026-54069) in SiYuan Note kernel's HTTP server allows any Chrome/Chromium browser extension to gain unauthenticated RoleAdministrator access, enabling data exfiltration, stored XSS injection, and configuration tampering for SiYuan desktop users, including via compromised legitimate extensions.

SiYuan publish-mode Reader can mutate Conf and SQL index via 8 ungated APIs, leading to configuration changes, denial of service, data corruption, and information disclosure by manipulating cloud sync intervals, graph configurations, SQL block content, and recent-documents lists.

SiYuan's Bazaar marketplace is vulnerable to stored cross-site scripting (XSS) via unescaped package metadata, leading to arbitrary OS command execution in the desktop Electron client.

A stored cross-site scripting (XSS) vulnerability exists in SiYuan due to the kernel storing Attribute View (AV) names without HTML escaping, allowing a malicious actor to inject arbitrary HTML which leads to Node.js code execution due to insecure Electron configuration, resulting in remote code execution (RCE).

SiYuan is vulnerable to path traversal via double URL encoding in the `/export/` endpoint, bypassing an incomplete fix for CVE-2026-30869; an authenticated attacker can exploit this vulnerability to traverse directories and read arbitrary workspace files, including the SQLite database (`siyuan.db`), kernel log, and user documents due to a redundant `url.PathUnescape()` call in `serveExport()`.

SiYuan is vulnerable to zero-click NTLM hash theft on Windows and blind SSRF on all platforms due to insecure Mermaid.js configuration, where a malicious Mermaid diagram containing a protocol-relative URL can be injected into a note, causing the Electron client to fetch the URL, triggering SMB authentication on Windows and sending the victim's NTLMv2 hash to the attacker. On macOS and Linux, the request acts as a tracking pixel and blind SSRF.

SiYuan Note versions prior to the fix for commit f09953afc57a are vulnerable to reflected cross-site scripting (XSS) via a namespace prefix bypass in the SanitizeSVG function when handling dynamic icons, allowing unauthenticated attackers to execute arbitrary JavaScript in a victim's browser.

SiYuan versions prior to 3.6.2 are vulnerable to remote code execution (RCE) via a malicious website exploiting a permissive CORS policy to inject a JavaScript snippet, leading to arbitrary code execution within the application's Node.js context.

SiYuan note taking application is vulnerable to a directory traversal via the /api/file/readDir endpoint, which does not require authentication, allowing an attacker to enumerate the directory structure and retrieve file names, potentially leading to arbitrary document reading.

SiYuan is vulnerable to arbitrary document reading via the publishing service, allowing attackers to retrieve document IDs and view the content of all documents, including encrypted or prohibited ones, by exploiting the `/api/file/readDir` and `/api/block/getChildBlocks` interfaces.

SiYuan versions 3.6.3 and below are vulnerable to arbitrary code execution due to insecure rendering of Mermaid diagrams, allowing injected javascript: URLs within Mermaid code blocks to execute arbitrary code when a victim opens a note containing a malicious Mermaid block and clicks the rendered diagram node.

SiYuan versions 3.6.3 and prior are vulnerable to path traversal (CVE-2026-40318), allowing attackers to delete arbitrary .json files on the server via the /api/av/removeUnusedAttributeView endpoint.

SiYuan versions 3.6.3 and below are vulnerable to unauthorized attribute view deletion via the /api/av/removeUnusedAttributeView endpoint, allowing authenticated users with publish-service RoleReader tokens to delete arbitrary attribute view definitions, leading to database view breakage and workspace rendering issues.