Zapscape KVM Flaw Lets Guest Code Escape Host

A new Linux kernel vulnerability named Zapscape has been identified by researchers at The Hacker News. The flaw allows an attacker who already holds kernel privileges within a Level‑1 (L1) guest virtual machine to break out of KVM isolation and execute arbitrary code directly on the underlying host system. It is tracked as CVE-2026-64561 and specifically targets KVM/x86’s shadow memory management unit (MMU), which manages the shadow page tables that separate guests from each other. ¹

The risk applies when nested virtualization is exposed to untrusted guests, meaning any cloud environment where a hypervisor runs VMs for users that are not fully trusted could be vulnerable. If a malicious VM gains kernel access it can read or write to the host’s memory through the MMU race condition, effectively bypassing the intended isolation boundary between guest and host. This type of attack was previously seen in Google Project Zero disclosures such as CVE‑2019–6345, but this is the first public report affecting modern KVM implementations. ¹

Because hypervisors form the foundation of most commercial cloud services—AWS, Azure, GCP, etc.—a single compromised guest could pivot laterally across multiple hosts sharing the same hypervisor infrastructure. In practice, an attacker inside one VM could gain control over the host running many other instances, potentially impacting thousands of users at once. The vulnerability echoes earlier rowhammer‑style attacks where memory access patterns were used for privilege escalation; here the exploit weaponises a race in the shadow page table rather than timing bit flips. ¹

Official records from NIST’s National Vulnerability Database show that the CVE ID assignment occurred on August 7th, with notification of the flaw made publicly by the researchers shortly thereafter. Because the issue directly undermines the isolation guarantees expected from nested virtualization, patching should be treated as urgent. Enterprises using KVM-based infrastructures are urged to update their kernels immediately and verify that no untrusted guests can obtain kernel privileges. ¹