Researchers recently uncovered a critical vulnerability that allowed them to execute their own code within a sealed portion of Windows—a Virtualization-Based Security (VBS) enclave meant to remain impervious even if the operating system is completely compromised. The flaw specifically affected the SmartScreen enclave component responsible for guarding sign-in credentials during phishing detection.
What Went Wrong: A Four-Byte Write & Timing Bug
The issue centers on a component named SFAPE.dll, which implements Enhanced Phishing Protection in SmartScreen. This enclave checks passwords by comparing what’s typed or pasted to stored hashed credentials, yet the vulnerability allowed attackers to subvert those protections. First, calling the enclave’s initialization function a second time enabled the attacker to write a fixed four-byte value to a memory address they control, bypassing the usual address validation. That write corrupted the mechanism that keeps the enclave’s memory separate from the host process.
Next, a timing vulnerability in another function was leveraged to trigger a controlled memory wipe. This wipe altered how the enclave looks up internal tables, allowing attackers to replace the structure that dictates which code can run inside the enclave. Once that foundation was in place, the exploit enabled malicious code to execute within the enclave’s trusted boundaries.
Scope, Impact, and Microsoft’s Fix
This vulnerability is tracked as CVE-2026-83498, rated Critical with a CVSS score of 7.8. Microsoft released a patch on September 8, 2026, in its cumulative update for Windows 11 versions 24H2 and 25H2. The fix adds address validation to the initialization function and strengthens the logic involved in the vulnerable timing path. Moreover, Microsoft’s policy update blocks loading the compromised enclave entirely—patched systems will be unable to use it.
It’s important to note the attack is local only. It requires prior low-level access to the machine—such as code running as SYSTEM—and cannot be carried out remotely or via common phishing vectors. While the proof-of-concept did not demonstrate theft of stored passwords, it did show potential for arbitrary code execution and data disclosure within a supposedly tamper-proof enclave.
Lingering Risks: Other Enclaves Remain Vulnerable
While the patch addresses this specific enclave, the broader enclave security model still has gaps. Windows currently lacks a global mechanism for blacklisting vulnerable enclaves—unlike its approach to malicious drivers. Microsoft’s policy update blocks only the old version of SFAPE.dll. Other signed enclaves with similar vulnerabilities, like a flawed one in Microsoft Edge marked by CVE-2023-36880, are still allowed to load and may be usable in future attacks.
If an attacker already has elevated access to a Windows system, this vulnerability allows them to further escalate trust boundaries, breaching protections designed to remain secure even under kernel control.
Why This Matters & What to Watch
Enclaves like VBS are foundational to modern Windows security, isolating sensitive functions—like password validation—from broader system compromise. When they’re compromised, all bets are off. Although Microsoft fixed the identified enclave, the existence of similar signed but flawed enclaves presents ongoing risk. Attackers could deploy “bring your own enclave” strategies by introducing any vulnerable enclave, signed or otherwise.
Moving forward, key things to monitor are whether Microsoft expands enclave blacklisting policies and how it strengthens design patterns to avoid similar timing or write-validation flaws. Users and administrators should ensure their devices are fully updated as of the September 2026 patch. For the longer term, this incident underscores the need for more rigorous auditing of enclave code, broader policy controls for signed components, and more consistent threat modeling.
