What this flaw actually does
A vulnerability tracked as CVE-2026-74580, named Vghost, was disclosed on October 12, 2026 with working exploit demos. It lives in the vhost subsystem of the Linux kernel, which is the part that handles fast network and I/O for virtual machines and containers. The short version: a local user with access to a device node can read and write arbitrary kernel memory, bypass KASLR, and become root. The attack is deterministic, meaning no lucky timing, no memory grooming. It just works.
If you’re running Linux servers, cloud VMs, or any virtualization infrastructure, this is the kind of flaw you patch before you do anything else this week. Affected kernels span from 4.11 all the way to 7.2-rc4, which is nearly a decade of releases. Tencent Zhuque Lab confirmed the full local-to-root exploit chain on Ubuntu 26.04, RHEL 9.8, Amazon Linux 2023, Microsoft Azure Linux 3.0, and Oracle Linux 10.2.
The two attacks you need to understand
The first attack is the local privilege escalation. Any user who can open /dev/vhost-net or /dev/vhost-vsock can exploit this. On four of the five tested distributions, those device nodes were world-readable at mode 0666, meaning any local account could reach them. Ubuntu restricted access to the kvm group, so there the attacker needs to be a kvm group member first, but that’s a lower bar than most people assume.
The second attack is narrower but more alarming for anyone running multi-tenant or shared virtualization. When a VM uses a virtio device behind a virtual IOMMU with iommu_platform enabled and vhost runs in device-IOTLB mode, an untrusted guest can trigger the same bug and make the host kernel write out of bounds into the host’s QEMU process. From there, the guest can overwrite a function pointer in QEMU and control where the host process executes next. That’s a VM escape.
The root cause is a cache invalidation gap that took two separate code changes to create. A 2017 change in Linux 4.11 added the cache and its shortcut. A 2020 change in Linux 5.9 moved address checking out of the command handler into a later step. When VHOST_SET_VRING_ADDR or VHOST_SET_VRING_NUM reconfigure a live ring, the cache isn’t cleared, so stale translations persist and the bounds check never runs.
What to do right now
Patch and reboot. The fix is commit de845981da67, which clears the cache whenever the ring is reconfigured. The first fixed upstream stable versions are 5.10.265, 5.15.216, 6.1.183, 6.6.152, 6.12.104, 6.18.45, and 7.1.9. Red Hat, Amazon Linux, and Oracle Linux had fixes shipped as of October 12. Ubuntu had fixes for 22.04 but as of October 13 listed 24.04 and 26.04 as still in progress.
While you’re waiting on a vendor patch, there’s a partial mitigation for the local privilege escalation path. Remove world access to /dev/vhost-net and /dev/vhost-vsock, and limit the owning group to trusted virtualization services only. Treat membership in that group as an actual privilege, not a default. That said, there is no interim workaround for the VM-escape path. The patched kernel is the only fix there.
Severity ratings vary. Kernel.org scores this at 8.8. Red Hat rates it 7.3. Amazon scored it 7.8. The disagreement is mostly about whether the guest-to-host write crosses a meaningful trust boundary. Practically speaking, a reliable local-to-root on production Linux servers is serious regardless of which score you use.
This flaw is a good reminder of why patching cadence on Linux infrastructure matters as much as it does on Windows. If you want a repeatable process for keeping Linux and mixed environments current, the same discipline applies as with any critical kernel-level patch response. And if the VM escape angle concerns you, the broader question of how your virtualization hosts are segmented and monitored is worth reviewing alongside your network security posture.
Check your kernel versions, apply available patches, restrict device node permissions where patches aren’t yet available, and verify your virtualization hosts are on the fixed list before assuming you’re covered. If keeping up with this across multiple servers feels like more than your team can absorb, booking an IT assessment is a reasonable next step.
FAQ
Does CVE-2026-74580 affect my cloud VMs on AWS or Azure?
Yes, potentially. Tencent confirmed the exploit on Amazon Linux 2023 and Microsoft Azure Linux 3.0. Amazon Linux has shipped a fix. Azure Linux fixes should be verified against your current kernel version. Check your running kernel and compare it against the fixed versions listed in the disclosure.
Can this flaw be exploited remotely over the internet?
No. CVE-2026-74580 requires local access to the system, specifically the ability to open a vhost device node. A remote attacker would need to compromise a local account first. That doesn’t make it low priority, but it does mean your internet-facing attack surface isn’t the immediate concern here.
What Linux kernel versions are vulnerable to Vghost?
Any kernel from 4.11 through 7.2-rc4 is affected. The flaw was introduced by a 2017 change in kernel 4.11 and compounded by a 2020 change in kernel 5.9. The first fixed stable releases include 5.10.265, 6.1.183, 6.6.152, and 6.12.104 among others.
Is there a workaround if I can’t patch immediately?
For the local privilege escalation, yes. Restrict access to /dev/vhost-net and /dev/vhost-vsock by removing world-readable permissions and limiting the device group to trusted services. For the VM-to-host escape path, there is no workaround. A patched kernel is required.
