A newly disclosed Linux kernel flaw lets any unprivileged local user escalate to root on most mainstream distributions, and a public exploit for it is already circulating. Tracked as CVE-2026-64531 and nicknamed OVSwrap, the bug sits in the Open vSwitch kernel datapath, the networking component that underpins a large share of cloud, container and virtualisation infrastructure.
The details come from a threat advisory published by Rewterz, which rates the issue high severity. The researcher who found it is not named in that advisory. The work is described as an experimental technique that combines large language models with structured memory-geometry visualisations to analyse complex kernel memory behaviour, which makes this one of the more concrete examples so far of machine assistance turning up a real kernel bug.
What's affected
The vulnerable code has been in place for roughly 13 years, but only became practically exploitable after a kernel change in 2025 removed an earlier size restriction. Exploitation was confirmed on Ubuntu, Debian, Fedora, AlmaLinux, Rocky Linux, Arch Linux, Kali Linux, Linux Mint, Amazon Linux, NixOS and openSUSE Tumbleweed, although some of those need minor configuration changes first. Older kernel branches that never took the 2025 change remain immune.
What should worry administrators is how low the bar is. An attacker needs no existing Open vSwitch bridge, no running OVS daemon, and no administrative privilege on the host. An unprivileged user can create isolated user and network namespaces with a tool as ordinary as unshare, pick up the CAP_NET_ADMIN capability inside that namespace, and stand up a private OVS datapath to reach the vulnerable code. Because Open vSwitch is commonly distributed as an auto-loadable kernel module, that attack surface can exist on machines where nobody has ever deliberately configured OVS at all. Rewterz also notes the same technique could potentially be adapted for container escape.
How the attack works
Open vSwitch receives network actions as nested Netlink attributes, the message format the kernel uses to talk to userspace. The kernel allows the overall action stream to exceed 64 KiB, but it failed to ensure each individual nested action stayed inside the 16-bit length field limit of 65,535 bytes. By crafting an oversized nested action, for example a CLONE containing a large number of conntrack actions, an attacker triggers an integer wraparound that resets the stored length to a much smaller number. The parser then resumes reading from attacker-controlled data instead of the legitimate end of the action, and forged actions are interpreted as valid. Attacker-controlled conntrack fields keep the offset predictable, so exploitation is highly reliable and does not need the delicate memory grooming that usually makes kernel bugs difficult to weaponise.
What you should do
The flaw was responsibly disclosed to the Linux kernel security team and the Open vSwitch maintainers, and fixes have landed in stable kernels 5.15.212, 6.1.178, 6.6.145, 6.12.97, 6.18.40 and 7.1.5. Upgrading is the primary mitigation, with distribution updates to follow as vendors ship them. Where patching has to wait, blacklist the openvswitch kernel module on hosts that do not need it, disable unprivileged user namespaces where that is operationally feasible, and consider the emergency BPF-based mitigation released alongside the public proof of concept.
Local privilege escalation tends to be filed under second-tier risk right up to the moment working exploit code exists, the same shift defenders faced when a public exploit appeared for a macOS printing flaw that writes files as root. Linux server estates have drawn steady attacker attention this year, including a run of critical Oracle, Kemp and Linux flaws under active attack.
This briefing is provided by IntelFusions for informational and defensive purposes only. It is based on sources assessed to be reliable at the time of writing, and analytic judgments carry the confidence levels indicated. Indicators of compromise are defanged; re-arm them only in controlled environments. IntelFusions is not affiliated with the organizations named and makes no warranty as to completeness or accuracy.