PwnEscape Linux Networking Flaw Exposes Host Memory From Guest VMs

Researchers disclosed PwnEscape (CVE-2026-99999), a critical Linux host-networking vulnerability that can let an attacker with code execution inside a guest VM disclose host memory and potentially compromise the host. Guest-controlled checksum metadata can cause deferred checksum processing to modify an IPv4 header after validation, with subsequent fragmentation enabling exploitation. Researcher Paulos Yibelo demonstrated host-memory disclosure against Vercel production sandboxes, while host crashes, selected kernel writes, and host-root execution were demonstrated on researcher-controlled systems. The reports describe repeatable 32-byte disclosures but no known malicious exploitation or demonstrated access to customers’ files, secrets, or application data. The flaw is in the host networking stack, not KVM itself.

Upstream commit 6785011 adds checksum-start validation across five networking paths, rejecting packets whose checksum start precedes the end of the IPv4 header. However, the patch is described as a partial defense-in-depth measure, with broader ingress validation still requiring work; the commit alone does not establish affected kernel versions or the full security impact. Operators should verify vendor advisories and stable-kernel backport status, update virtualization hosts to vendor-confirmed patched builds, and reboot into the updated kernels. Updating guest kernels alone does not remediate the host vulnerability. AWS also lists the CVE, but the supplied reference provides no package or remediation details.