A Sandbox Escape Without a Single Memory Corruption Flaw

Most browser exploits that make headlines rely on memory corruption: a use-after-free here, an out-of-bounds write there, stitched together into a chain that climbs from a script running in a web page to code executing on the underlying operating system. Orange Tsai of DEVCORE has demonstrated that this assumption is not a requirement. In a full technical write-up published on October 28, the researcher detailed how a chain built entirely from logic flaws was enough to break out of Microsoft Edge's sandbox on Windows and land code execution on the host machine.

The chain earned $175,000 at the Pwn2Own Berlin contest in May 2026, and the disclosure arrived roughly five months later. According to The Hacker News, Microsoft patched the affected Edge bugs the day after the contest concluded.

Why Logic Flaws Change the Economics of Exploitation

Memory corruption bugs are fragile. They depend on heap layouts, allocator behavior, and precise timing, and they frequently break when a vendor ships a minor update. Logic flaws behave differently. They exploit incorrect assumptions in how software validates input, checks permissions, or decides which component is allowed to talk to which. Because they stem from design and implementation mistakes rather than memory-safety failures, they can survive code changes that would neutralize a traditional exploit, and they can sometimes be reproduced with far more reliability.

The significance of Tsai's work is not simply that a logic-only chain exists. It is that the chain was sufficient to cross a security boundary that Microsoft has invested heavily in hardening. Edge's sandbox is designed on the premise that a compromised renderer should be contained, with the broker process acting as the gatekeeper for privileged operations. A chain made of logic bugs suggests the gatekeeper can be talked past rather than smashed through.

The Attack Surface: Browsers as the Front Door

For defenders, the practical takeaway begins with the threat model. A malicious web page remains one of the most accessible delivery mechanisms in existence: no attachment to open, no file to download, no user training to defeat beyond convincing someone to visit a link. When the exploitation primitive is a logic error rather than a memory-safety defect, exploit mitigation technologies such as Control Flow Guard, Arbitrary Code Guard, and various heap hardening measures offer little protection, because the attacker is not corrupting memory in the first place.

This does not mean memory-safety defenses are wasted effort. It means they are incomplete. Organizations that treat browser isolation, sandboxing, and exploit mitigations as a complete answer to drive-by compromise should revisit that assumption.

What Security Teams Should Do

Several concrete actions follow from this research.

  • Prioritize browser patching above almost everything else. Microsoft fixed these issues within a day of the contest. The window between a working exploit and a fix was measured in hours, but the window between a fix and enterprise deployment is often measured in weeks. Browser updates should be treated as emergency change management, not routine maintenance.
  • Assume the browser is a hostile execution environment. Where feasible, remote browser isolation, application containment, and strict separation between browsing activity and sensitive workloads reduce the blast radius if a chain like this is used in the wild.
  • Invest in logic-flaw testing. Threat modeling and manual review should explicitly ask where trust boundaries are enforced by checks that could be bypassed through sequence manipulation, state confusion, or confused-deputy patterns, rather than only where memory could be corrupted.
  • Watch for chained exploitation in telemetry. A logic-only escape may not trigger the crash-based detections that security operations centers have tuned for. Behavioral monitoring of browser child processes, unusual inter-process communication, and anomalous process creation remains valuable.

The Broader Lesson

Pwn2Own results are often read as scoreboards, but their real value is as early warning about where the security industry's assumptions are weakest. Tsai's chain is a reminder that the most elegant attacks are frequently the ones that avoid the fight entirely: instead of defeating a mitigation, they find the path where the mitigation was never asked to apply. For browser vendors, that means auditing the logic of privilege transitions with the same rigor applied to memory safety. For everyone else, it means accepting that the browser is a full-fledged attack surface and defending it accordingly.

The full technical details are available in the researcher's published write-up, linked through the original report.