CVE-2026-72789

This vulnerability exposes a dangerous interaction between two features that were designed in isolation: SiYuan's notebook encryption and its publishing system. When you encrypt a notebook, the system gives you no indication that publishing it is even possible—yet the publish API will happily enumerate encrypted notebooks as available to anonymous readers. This isn't a buffer overflow or injection flaw. It's a compositional failure: encryption was implemented as if it created a terminal security boundary, while publishing was built on the assumption that anything it finds is intentionally meant for public consumption. The enumeration vector is particularly insidious because it's active rather than passive. A user who monitors their published content will see their encrypted notebooks listed there, creating immediate cognitive dissonance with their security intent. The exposure window is also temporal: encrypted notebooks become visible during the unlock-edit-lock cycle when a user is actively working. Polling the publish API during that window reveals content the user believes is protected. This pattern isn't unique to SiYuan. The same failure appeared in Evernote (2019), Notion (2021), and Confluence (2022). The recurring issue is treating encryption as a lockbox rather than a sensitivity classification—the system loses track of data sensitivity once it's decrypted for authorized use, and subsequent features like publishing, sharing, or webhooks never know they're handling protected content. Check your SiYuan instances immediately. Verify which notebooks are configured for publishing and confirm none of them are or were encrypted. Review server logs for anonymous access patterns to the publish endpoint—without an audit trail, you may have no way to know if this was exploited. The v3.7.4 patch reportedly addresses this, but examine whether it creates a hard block or just adds a warning that users can bypass. If it's the latter, the underlying architectural assumption remains unfixed and future features will likely reproduce this vulnerability.

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