A significant vulnerability has been identified in the Linux kernel, enabling virtual machines to access the memory of their host systems. This flaw, dubbed PwnEscape and officially tracked as CVE-2026-99999, was discovered during a security challenge by Vercel and was later confirmed by the company. Notably, the vulnerability is located within the Linux kernel’s networking code, affecting the foundational layer used by many cloud providers.
Details of the PwnEscape Vulnerability
Security researcher Paulos Yibelo from pwn.ai found that specific network packets sent from a virtual machine could exploit the flaw to read host memory, crash the host, or execute code on it. These findings were verified in Vercel’s live cloud environments, revealing that each packet sent could return approximately 32 bytes of the host’s memory. Although the memory leak was observed in production environments, the crash and code execution were only tested on systems set up to mimic Vercel’s configurations.
Impact and Exploitability
The vulnerability has been rated 9.3 out of 10, classified as critical, by the kernel maintainers. The U.S. National Vulnerability Database is aware of the issue, though it hasn’t finalized its review. The flaw requires an attacker to have local access to the system, as demonstrated by the researcher, and cannot be exploited remotely over the internet. Currently, there are no reports of this vulnerability being actively exploited in real-world scenarios.
Mitigation and Patch Details
Red Hat’s Michael S. Tsirkin developed a fix integrated into the Linux kernel, which introduces an additional check in the networking code. The fix is available in several stable kernel branches, including versions 5.10.272, 5.15.223, and others. Users are advised to update their host kernels to these versions or consult their distribution for specific patches. While the update addresses the vulnerability, it is recommended to disable nested virtualization and apply restrictive permissions to limit potential damage if exploited.
This vulnerability was identified during a two-week challenge hosted by Vercel, offering up to $1 million for successful sandbox escapes. The discovery was accomplished using pwn.ai’s AI-based research tools, which analyzed source code and directed the investigation, demonstrating the growing role of AI in cybersecurity research.
