SiYuan CVE-2026-72804

Exploitation requires a SiYuan instance (before v3.7.4) that is network-reachable and serving content through its publish mode, containing documents protected by a publish-password; the attacker then calls the getGraph or getLocalGraph endpoints directly without supplying that password. No authentication, credentials, or user interaction are needed (PR:N/UI:N), and default publish deployments are vulnerable. The practical limiting factors are that the instance must be exposed to the attacker's network position and must actually be running publish mode with password-protected documents - a purely local, single-user desktop instance with no published surface is not exposed, and only confidentiality is at risk (no write or denial-of-service impact).

Signals are consistent and point to a genuine, easy-to-exploit confidentiality issue: the CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:N, VC:H) describes a network-reachable, low-complexity, no-privilege, no-interaction attack, matching the description of anonymous readers bypassing the password tier. The impact is bounded to confidentiality - there is no integrity or availability effect (VI:N/VA:N) - so this is data exposure, not takeover, which tempers the 9.2 base score in practical terms. There is no EPSS value, no KEV listing, and no POC in the provided data, so likelihood of opportunistic mass exploitation is unproven and any operator running SiYuan in publish mode with protected documents should still treat exposure of confidential notes as urgent and patch promptly. The main caveat: real-world risk hinges on whether the instance is actually reachable and running publish mode, which the CVE data does not confirm per-deployment.

An attacker who can reach a SiYuan instance published on the network sends a direct request to getGraph or getLocalGraph with no publish password. The server returns the block-level content and reference links of documents the owner believed were password-protected, letting the attacker reconstruct confidential notes and their interconnections. No exploit code is publicly identified, but the attack requires only a crafted HTTP request and no authentication or user interaction.