A KVM guest escape that runs code on your hypervisor
Zapscape is a guest-to-host escape for KVM on x86. It lets a compromised VM run code on the host machine, which is the nightmare scenario for cloud providers.
Someone published Zapscape, a working exploit for CVE-2026-64561 that breaks out of a KVM virtual machine and runs code on the host. This is the exact scenario AWS and GCP spend millions preventing. The bug is in the x86 KVM implementation. The exploit chain targets the memory management subsystem where the hypervisor translates guest physical addresses to host physical addresses. Get that wrong and a guest can write outside its sandbox. What makes this interesting is the timing. The CVE was assigned in 2026, meaning this is either a very recent disclosure or the embargo just lifted. Either way, cloud providers are patching right now. If you run KVM in production, this is the week to check your kernel version. The repo includes a proof-of-concept that spawns a shell on the host from inside the guest. That is not a theoretical risk. That is a working demo. The hypervisor is supposed to be the trust boundary. When a cloud provider sells you a VM, the promise is that your neighbour cannot read your memory or touch your disk. A guest-to-host escape breaks that promise. One compromised VM and the attacker owns the physical server, which means they own every other VM on that box. This is why kernel security researchers get paid. One bug in the KVM memory subsystem and the entire multi-tenant cloud model collapses. Zapscape is a reminder that virtualisation is a software problem, and software has bugs.
Source: Zapscape (CVE-2026-64561): Guest-to-Host Escape in KVM/x86