Executive brief
Jupyter Server, the core engine for Jupyter web applications, contains a security flaw that allows users to access files they should not be able to see. An authenticated user can bypass directory restrictions to read, modify, or delete files in other users' folders if those folders share a similar name. This is particularly dangerous in shared environments where many users have predictable folder names, potentially leading to data theft or loss.
Technical details
A path traversal vulnerability exists in the Jupyter Server REST API (/api/contents endpoint) due to improper validation of encoded path components. Authenticated attackers can escape the configured 'root_dir' to access sibling directories that share the same naming prefix (e.g., a user with access to 'user1' could access 'user10'). The exploit involves sending a crafted POST request to the checkpoints sub-endpoint using '%2e%2e/' sequences. This allows for full unauthorized file system operations (read/write/delete) within the scope of the prefixed sibling directories. The issue is fixed in version 2.18.0.
Affected products
- Jupyter Jupyter Server <= 2.17.0
- Red Hat Migration Toolkit for Applications 8 8
- Red Hat Red Hat OpenShift AI (RHOAI)
Timeline
- 2026-05-05: disclosed
- 2026-05-05: advisory
- 2026-05-05: patched: Fixed in version 2.18.0