Microsoft Edge: a logical chain from UXSS to code execution

A researcher from DEVCORE presented a complete Microsoft Edge exploit chain built entirely from logic bugs. The chain consists of five bugs and two Windows features. It enables an attacker to go from JavaScript execution on an arbitrary website to arbitrary code execution on the system.
How it all started: Navigation Throttles The research began with the Navigation Throttles mechanism in the browser process. One of them turned out to be EdgeCAGuidanceNavigationThrottle — a component of Edge's enterprise Entra ID feature. Its purpose is to guide users to the appropriate profile when Edge detects them signing in with a work account on a Microsoft website. Once triggered, the throttle cancels the original navigation and launches a new one using the address specified in switch_profile. Since this new navigation is initiated by the browser process, it can navigate to pages normally inaccessible to JavaScript, including edge:// and javascript: URLs.
Four obstacles and how they were bypassed Turning UXSS primitive into an exploit required solving four challenges.
  1. Trusted origin. The throttle checks the origin using GetLastCommittedURL() of the current tab, but navigation can also be initiated from another window via window.open.
  2. Pop-up blocker. Microsoft implemented its own allowlist for pop-ups in Edge. A window opened with about:blank#quickAuthPopup is allowed, after which it can be redirected to any website.
  3. Requirement for a signed-in profile. The OneAuth feature automatically signs users in to their Microsoft accounts when visiting Microsoft websites. Starting with Windows11 25H2, signing in to Windows requires an MSA account (unless a local-account workaround is used). Even without an Edge profile, simply visiting microsoft.com is enough for Edge to create one and sign in automatically.
  4. Requirement for prior knowledge of the login email. The assets.msn.com website, embedded in Edge's new tab page, returns the currently signed-in user's email address. Due to a CORS misconfiguration, it can be retrieved from any website using fetch with credentials.
Escaping the browser sandbox UXSS alone does not provide code execution. Privileged pages such as edge://settings are protected: Chromium ignores javascript: URLs on them. However, Edge has its own Reading Mode called Immersive Reader, which uses a custom read:// scheme, which Microsoft did not add to the blacklist. This privileged WebUI provides access to the chrome.edgeFeedbackPrivate.zipDiagnosticLogFiles() function. It accepts a filename and a JSON string, writes the data to a temporary folder, and packages it into an archive. Exploiting path traversal (../) and a null byte in the filename turns it into arbitrary file write primitive primitive with control over both the filename and contents.
From arbitrary file write to code execution Instead of of DLL sideloading, which was hindered by the JSON format, the researcher overwrote protocols.json, which Edge reads during initialization. After restarting via edge://restart, a protocol handler such as telnet:// can be launched from any origin. The handler looks for telnet.exe in the directories listed in the PATH environment variable, and one of them — C:\Users\<user>\AppData\Local\Microsoft\WindowsApps — is writable by a low-privileged user. Uploading a modified executable there is enough to complete the chain and achieve arbitrary code execution.
Vendors
Products
More