Security researcher Orange Tsai has published a technical account of the Microsoft Edge exploit chain he developed for Pwn2Own Berlin 2026, describing a route from browser navigation behavior to a sandbox escape using logic flaws rather than memory-corruption bugs. The disclosure was posted on October 29 after the work had also been presented at HITCON and Hexacon.
Tsai, a member of DEVCORE’s research team, says the competition chain earned $175,000 and 17.5 Master of Pwn points. His account describes five bugs combined with two Windows features. The supplied evidence is the researcher’s own write-up, so the implementation details and characterization of the chain are attributed to him.
The investigation began in Edge’s handling of Chromium “navigation throttles,” browser-process components that can intercept requests. Tsai focused on an Edge-specific throttle associated with Microsoft Entra ID conditional access. It is intended to help a user switch to the correct browser profile when signing into a Microsoft service with a work account.
According to the write-up, the feature could take a destination from an attacker-controlled query parameter and start a fresh navigation from the browser process. The first important weakness emerged from ambiguity over which browser tab supplied the navigation context. When no source tab was explicitly identified, Chromium could use the currently active tab. By changing tabs during a delay, the researchers say they turned that behavior into universal cross-site scripting, or UXSS, allowing code to execute in another site’s context.
Tsai then describes overcoming restrictions that initially appeared to limit the technique to designated Microsoft origins. The throttle checked the last committed address of a page, but a navigation could originate through another window. A cross-window request therefore offered a way around the origin assumption. Popup blocking created another obstacle, leading the team to examine Edge-specific popup allowances as part of the broader chain.
The work is notable because it targets trust and state-management errors rather than unsafe memory operations. Modern browsers isolate web content through multiple processes, origin rules and sandbox boundaries. A logic chain attempts to make individually legitimate components act on the wrong assumptions, progressively gaining capabilities that no single step was designed to provide.
The published account also emphasizes manual code review. Tsai says he examined a comparatively little-studied navigation area, traced browser and renderer interactions, and assembled the exploit without AI assistance. That framing is personal commentary, but the technical lesson is broader: mature codebases can retain security-critical gaps where product-specific features meet shared browser infrastructure.
Users and administrators should continue applying Edge and Windows security updates promptly. The write-up is valuable for defenders because it identifies a class of failure—missing context across navigation boundaries—that reviewers can seek in other browser features, without requiring a memory-safety flaw to become serious.
