What happened
Published 2026-10-13 (CVSS 4.2, CWE-367/863) via GitHub advisory. Verified via NVD REST API.
Why it matters
Notebook trust is the boundary that decides whether untrusted AI/ML notebook cells can execute; a trust-misattribution race undermines that decision inside the primary data-science execution environment.
Attack vector
The trust command-link handler resolves tracker.currentWidget instead of retaining the clicked document; if the user switches documents while the prompt is open and answers Trust, the second document gets trust-marked (all cells trusted, unsaved edits saved, server signature persisted), and later command buttons run without prompting.
Affected systems
JupyterLab 4.5.7–4.5.11 / 4.6.0–4.6.4; notebook 7.5.6–7.6.3; jupyterlite-core 0.7.5–0.8.4
Mitigation
Upgrade JupyterLab to 4.5.11 / 4.6.4 (and affected Notebook/JupyterLite builds) per GHSA-jwrc-gm9j-263p.