Vulnerabilities & Exploits · Zero-Day Exploit

PwnEscape breaks Linux VM isolation at the host layer

A researcher privately reported CVE-2026-99999, nicknamed PwnEscape, after showing that a guest VM could read host memory through the Linux kernel's networking code. The leak was demonstrated against Vercel's live sandboxes, and Vercel confirmed the finding; host crash and code execution were reproduced only in lab systems, with no sign of in-the-wild abuse.

The bug sits in shared kernel networking, so a specially shaped packet from inside a VM can make the host hand back pieces of its own memory instead of isolating the tenant cleanly. In the lab, the same flaw could also crash the host and break out of the VM, which means the failure point is the host kernel rather than any one cloud's sandbox wrapper.

That makes this a provider-layer problem, not a tenant one: if a platform runs untrusted code in Linux-backed VMs or sandboxes, the blast radius is every workload on the affected host. The reporting does not say which fleets are exposed yet, only that the isolation boundary itself is what fails.

1 source · 10h ago

CVE-2026-99999

NVD KEV

CVSS 9.3 CRITICAL: in the Linux kernel, the following vulnerability has been resolved: ipv4: validate checksum_start before completing…

Timeline

Sources

Part of the PlainSec briefing for 2026-10-16

Every edition of this story: PwnEscape breaks Linux VM isolation at the host layer

More from today