SiYuan versions before 3.6.4 configure Mermaid.js with securityLevel set to "loose" and htmlLabels enabled, allowing <img> tags with src attributes to survive internal sanitization and be injected into SVG <foreignObject> elements via innerHTML without secondary sanitization. This enables an attacker to craft malicious Mermaid diagrams that cause the Electron client to fetch attacker-controlled URLs. On Windows, protocol-relative URLs (e.g., //attacker.com/image.png) are resolved as UNC paths (\\attacker.com\image.png), triggering automatic SMB authentication attempts that leak the victim's NTLMv2 hash to the attacker. The vulnerability is classified as CWE-918 (SSRF) and has a CVSS 4.0 score of 8.7 (high severity). The issue is fixed in SiYuan version 3.6.4.

An attacker can craft malicious Mermaid diagrams embedded in notes that cause the victim's Electron client to make unintended network requests. On Windows systems, this can lead to automatic SMB authentication attempts that expose the victim's NTLMv2 hash to the attacker, potentially enabling credential theft or relay attacks. The vulnerability affects all versions of SiYuan prior to 3.6.4. There are no known exploits in the wild at this time.

This vulnerability is fixed in SiYuan version 3.6.4. Users should upgrade to version 3.6.4 or later to remediate this issue. No additional mitigations are specified. Patch status is not explicitly stated beyond the fix in 3.6.4; users should verify upgrade availability from the vendor.