A sophisticated chain of logic flaws, meticulously crafted by **Orange Tsai** of **DEVCORE**, has demonstrated how a malicious webpage can break free from **Microsoft Edge**'s sandbox and execute code on a Windows computer. Remarkably, this exploit achieved full control without a single instance of memory corruption. Tsai publicly detailed the full method on October 28, five months after his chain secured a substantial $175,000 prize at the **Pwn2Own Berlin** contest in May. **Microsoft** promptly addressed the identified bugs the day following the competition, ensuring that up-to-date browsers are no longer vulnerable. Browser sandboxes are critical security mechanisms designed to isolate web page code, preventing it from interacting with the underlying operating system. Bypassing this sandbox and subsequently executing arbitrary programs is the hallmark of a severe browser exploit.  ### The Novelty of Logic Flaws Historically, browser escapes almost invariably hinge on memory corruption vulnerabilities. Tsai's breakthrough, however, leverages a series of logic errors in how **Edge** and **Windows** handle ordinary features. His write-up reveals an exploit chain under 100 lines of code, showcasing the elegance and potency of this approach. This chain ultimately results in a universal cross-site scripting (UXSS) primitive, enabling the execution of JavaScript on any website, effectively bypassing Content Security Policies (CSPs). Crucially, because the attack originates from the browser's own process, it can target domains like `google.com` that are typically beyond the reach of standard web bugs. Tsai emphasizes this as a new attack surface rather than an isolated vulnerability. ### Competition Exploit, Not In-the-Wild Attack It's important to note that this exploit was a competition demonstration, not an attack observed in the wild. Tsai performed the exploit on stage at **Pwn2Own**, and there is no evidence of its use against real users. Public **CVE** records list exploitation as 'none.' One step in the chain involves restarting the browser, which Tsai clarified was a convenience for the contest and not essential for a real-world attack. ### Unpacking the Exploit Chain Tsai's chain begins with **Edge**'s conditional access navigation throttle, a feature designed for business sign-ins. He discovered that triggering this feature allows the browser's own process to initiate a new navigation untethered from the originating page. **Edge** then erroneously attaches this navigation to any open tab, creating the initial bug and the UXSS primitive. Achieving this without user interaction required exploiting four additional weaknesses: * **Trusted Site Bypass:** **Edge**'s check for trusted **Microsoft** sites was bypassed by opening the request from a second window. * **Popup Blocker Evasion:** **Edge**'s popup blocker failed due to a hardcoded list of allowed addresses, one of which facilitated the popup. * **Auto-Sign-in Leverage:** The attack capitalized on **Edge**'s auto-sign-in feature when a user visits a **Microsoft** site, ensuring the victim was already authenticated. * **Cross-Origin Misconfiguration:** The victim's email address was read from `assets.msn.com`, which returned it to any website due to a misconfigured cross-origin setting. ### Sandbox Escape and Code Execution Even with universal JavaScript execution, control over the computer remains elusive. To escape the sandbox, Tsai exploited **Edge**'s Immersive Reader, which operates under a custom `read://` address. **Microsoft** had inadvertently omitted this address from the list of pages barred from running scripts. From within Immersive Reader, the chain invoked an internal **Edge** function intended for zipping diagnostic logs. This function was then abused to write a file with an attacker-chosen name and content anywhere on the disk. File writing was escalated to code execution in two stages: 1. **Protocol Handler Overwrite:** The chain overwrote a configuration file that dictates which websites can launch specific protocol handlers, then restarted **Edge** to load the modified file. 2. **Telnet Exploitation:** A `telnet://` link was launched. Since **Windows** doesn't ship with a telnet program by default, the system searches its path for one. The attacker planted a malicious `telnet` executable in a low-privileged user-writable folder, which **Windows** then executed. ### The Fix and Industry Recognition **Microsoft** released the fix on May 15, the day after the contest, in **Edge Stable version 148.0.3967.70**. This update included three **Edge**-specific fixes: **CVE-2026-45492** (security feature bypass), **CVE-2026-45494** (spoofing bug), and **CVE-2026-45495**. These affect **Edge** versions prior to `148.0.3967.70`. **Microsoft** rated the first two **CVEs** at 5.4 on a 10-point scale, a medium score reflecting individual bugs, not the full severity of the chained exploit. **DEVCORE** confirmed the fix was deployed within 24 hours of disclosure. Users with automatic updates enabled for **Edge** would have received the patch in May, requiring no further action. Tsai's two-month, AI-free endeavor to construct this exploit chain underscores the enduring importance of human intuition and craft in uncovering complex vulnerabilities. His research earned the 2026 **Pwnie Award for Epic Achievement**.