Linux PwnEscape flaw lets guest VMs read host memory
A Linux kernel networking flaw, dubbed PwnEscape (CVE-2026-99999), let code inside a virtual machine read host memory in a demonstration against Vercel’s production sandboxes, the company confirmed. Researcher Paulos Yibelo demonstrated host crashes and code execution only on his own test machines; no real-world exploitation has been reported, and Linux maintainers have released a fix for host kernels.
Bottom line — The fix is in stable kernel releases from 5.10.272 onwards; updating a guest kernel alone does not protect the host.
Go deeper 6
-
The Hacker News reports that each crafted packet could disclose about 32 bytes of host memory, while Vercel said no challenge report showed access to customer files, secrets or application data.
-
The flaw requires an attacker to be running code inside a guest VM; it is not directly exploitable by an outsider sending packets over the internet, according to The Hacker News.
-
pwn.ai says its lab tests progressed from memory disclosure to selected kernel writes and host-root code execution. It did not claim code execution in a cloud provider’s production environment.
-
Red Hat’s Michael S. Tsirkin wrote the upstream fix, commit 6785011f8b16. The Hacker News says it adds checks at five points and is described by its author as a defence-in-depth measure.
-
The Hacker News lists fixed releases for stable branches: 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.57 and 7.2.11. Distribution vendors may include the fix under different version numbers.
-
Vercel’s August sandbox challenge drew 1,285 reports and paid out $242,000 (€222,000); The Hacker News says PwnEscape was its only Critical finding.