Bypassing Windows Administrator Protection via UI Access Token Abuse
What Happened — Google Project Zero disclosed a set of nine bypasses of the newly‑released Windows “Administrator Protection” feature. The research focused on five of those bypasses, all of which stem from the way the operating system implements the UI Access token flag. By setting this flag, a low‑integrity process can sidestep the User Interface Privacy Isolation (UIPI) checks that normally block a medium‑integrity process from interacting with high‑integrity windows. The technique allows a limited user to send window messages, install hooks, or otherwise manipulate privileged UI elements, effectively elevating privileges. Microsoft has confirmed the findings and released patches that remove the unsafe token handling paths.
Why It Matters for Trust & Control Assurance — The flaw demonstrates how a mis‑implemented access‑control exception can undermine the boundary that “Administrator Protection” is meant to enforce, highlighting the need for continuous verification of privileged‑access controls.
Who Is Affected — All organizations running affected Windows versions (Windows 10 1909 and later, Windows 11) across any industry that rely on UAC and UI Access for accessibility tools or legacy applications.
Recommended Actions
- Deploy Microsoft’s latest cumulative updates that address the UI Access bypasses.
- Inventory any third‑party applications that request the UI Access flag and validate their necessity.
- Review and tighten group‑policy settings that allow UI Access for non‑essential processes.
- Incorporate privileged‑access control testing into your regular audit‑readiness cycles.
Source: Google Project Zero blog
Technical Notes
- Attack vector: Abuse of the UI Access token (privilege‑escalation via UI message injection).
- Root cause: UI Access flag granted blanket bypass of UIPI, originally added for accessibility scenarios.
- Mitigation: Microsoft patched the token handling logic and tightened the conditions under which UI Access is honored.
- Data types: Not applicable – the issue is a privilege‑escalation pathway, not a data‑leak.
Source: same as above