Details of the Linux kernel vulnerability “PwnEscape” (CVE-2026-99999) were disclosed in October 2026. An attacker on a virtual machine could potentially execute malicious code on the Linux OS running on the host outside it. The vulnerability has a severity score of 9.3 on a scale of 1 to 10, placing it in the top “Critical” category.
A crack in the “wall” of virtual machines
Virtual machines let users create multiple isolated environments on a single physical server. Many VPS, hosting and cloud services rely on this technology. We use them with the assurance that other users’ environments are invisible and inaccessible to us.
PwnEscape targets that wall itself. If an attacker can escape a guest virtual machine and reach the host’s Linux system, other users sharing the same physical server—and even the provider’s own systems—could be put at risk. It is like using the key to your apartment to get into the building’s management office.
A gap in network-traffic inspection provides an entry point
A technical explanation was published in an email to Openwall, a mailing list for security specialists. The cause lies in the Linux kernel’s network processing. The explanation says that data related to the “checksum”—used to check whether network data has been corrupted—could be modified by an attacker.
Normally, once the kernel has validated an IPv4 header—the part containing information such as the destination—it trusts the header and continues processing. But this flaw made it possible to overwrite a validated header in place of the checksum. Using this as a foothold, an attacker could disrupt normal network processing and read from and write to memory locations that should be inaccessible. In a research environment, the exploit reportedly went as far as executing code on the host Linux OS.
The email’s subject line reads “Guest-to-Host KVM Escape,” pointing to KVM, Linux’s virtualization technology. However, because the underlying cause lies in network processing, a core part of the kernel, the flaw may also be exploitable in environments other than KVM, such as containers.
What has been demonstrated—and the attacks that could follow
So far, the vulnerability has been demonstrated only in a research environment. This does not mean real-world attacks are currently spreading.
The email suggests that the security industry will use this information to investigate the impact on other hypervisors—the software platforms that run virtual machines. It also predicts that exploit code targeting configurations with weak defenses could appear within weeks, followed by code that works against hypervisors in their default configurations within a month.
These are forecasts, not established facts. But now that technical details are public, it is sensible to act on the assumption that there is little time to spare.
Users should look at how their providers are responding
Providers offering Linux virtual machines as a service are particularly pressed to act quickly. A malicious user could escape from their own virtual machine and attack other users or the provider’s systems.
As of the afternoon of October 17, 2026, Sakura Internet in Japan announced that it had stopped accepting new sign-ups for some services, including Sakura VPS and Sakura Rental Server, and had begun taking countermeasures. AWS says its managed compute and container services, including Amazon EC2, AWS Lambda and AWS Fargate, are not affected and require no action from customers.
Even when services are all described as “Linux virtual machines,” the impact can vary by provider and service model. Businesses and individuals using VPS, hosting or cloud services should not assume they are safe based on the service name alone. It is wise to check what services the provider says are affected and what measures it has taken. Virtualization security depends not only on users’ configurations but also on the actions of the operators running the underlying infrastructure.
Comments