CVE-2026-72799

This vulnerability is a trust boundary violation dressed as a utility feature. Five endpoints in SiYuan's publish mode — those resolving paths to IDs, navigating hierarchies, resolving titles, and enumerating siblings — allow anonymous callers to reconstruct the complete topology of documents that users explicitly marked as hidden, password-protected, or publish-forbidden. The attack surface isn't the document content itself; it's the map of what exists. That distinction matters because CVSS 5.8 captures technical severity but ignores strategic value: an attacker who enumerates your notebook names, folder hierarchies, and document IDs gains the targeting system for every downstream operation. Those IDs become input for privilege escalation, content exfiltration, and targeted phishing. You don't just learn that documents exist — you acquire resolvable handles that bypass the access controls those documents rely on. The core gap is between what the UI communicates and what the API enforces. When a user marks a document hidden, the interface tells them it's excluded from the publish surface. These path-resolution endpoints treat that intent as advisory. The attacker operates in the space between explicit user decision and actual enforcement — and that discrepancy is where impact assessment should focus, because it determines whether affected users can even identify their exposure. The structural pattern here is well-established across knowledge management systems. Confluence, Notion, MediaWiki, and GitLab wikis have all produced CVEs with nearly identical descriptions: authenticated-only path-resolvers that expose document topology when placed behind anonymous-accessible publish surfaces. The common thread is architectural: publish modes are typically layered onto existing authenticated systems, and the path-resolution infrastructure predates the threat model that publish mode introduces. Developers building internal workflows assume authenticated context; the publish layer arrives later and inherits infrastructure that was never audited against anonymous access. The v3.7.4 patch fixes these five endpoints, but unless the audit extends to the entire path-resolution surface in publish mode, the structural precedent suggests more instances will emerge. For detection and verification, note the asymmetry: defenders must verify all five endpoints are blocked, but an attacker needs only one working to reconstruct the graph. Legitimate enumeration behavior should be sparse and bounded — a user resolving a few known paths. Sustained, systematic queries across multiple endpoints, cycling through notebooks and folders, is the signature of an attack. The IDs are persistent, meaning the exposure compounds over time: each document a user creates after publication is immediately enumerable. The window between topology obtained and targeted content accessed is where the actual risk materializes, and that window extends across the patch-to-deploy lag. Prioritize remediation not because the individual endpoint is severe, but because the intelligence it provides enables every subsequent operation against the publish surface.

Reviewed through automated stages and approved by a human before publication.