# MPTCP opens CVE-2026-98432 and Linux remains unpatched on RHEL

> The CVE-2026-98432 exploit targets MPTCP listeners, which Go 1.24 opens by default. Debian has already patched, while Ubuntu and RHEL still have no firm date.
> By José Maldonado · Cybersecurity · 2026-11-02

The most discussed Linux vulnerability of the month has a specific door and almost no one has named it. The exploit that an Anthropic research model turned into a remote root shell targets MPTCP listeners, the transport protocol extension that allows several network paths to be used at once. And there the detail appears that changes the risk calculation for thousands of production services: since Go 1.24, the runtime enables MPTCP by default on listening sockets.

The flaw, catalogued as CVE-2026-98432, is a use-after-free in the IPv4 TCP code. The kernel CVE team scored it 9.8 out of 10 on October 24, after the patch reached the stable branches on October 5 and the public record was published on the 15th. It affects kernels from version 4.4 onward, which means the bug has been inside the code for eleven years without a human audit finding it. The full record of the flaw, with affected versions and status by distribution, is in [the tracking we published on October 31](https://observatorioblockchain.com/en/cybersecurity/cve-2026-98432-kernel-tcp-ubuntu-rhel-sin-parche/).

The patching status remains uneven three weeks after the discovery. Debian published the DSA-6568-1 update on October 16 with linux 6.12.113-1 for Trixie and versions 6.1.190-1 and 6.12.113-1~deb12u1 for Bookworm. SUSE, Oracle and Amazon Linux fixed it in their supported branches. Ubuntu has the kernels in preparation and Red Hat has not committed to a date for RHEL 8, 9 or 10.

## What is MPTCP and why does it change the real risk?

MPTCP allows a TCP connection to use several network interfaces at the same time, so that if one path fails the traffic continues over another. On mobile devices and in deployments with redundant links it is an obvious advantage. On Linux servers it has been little used because it adds complexity to connection tracking and because most applications do not need it.

The problem is that the MPTCP code shares structures with the normal TCP handshake, and the use-after-free is triggered during that handshake. A listener that accepts MPTCP connections exposes a wider surface than a classic TCP listener, with more intermediate states to free. The practical consequence is that the exploit has something to lean on in services that until now were considered little exposed.

That is where Go comes in. Since version 1.24, the standard library enables MPTCP on listening sockets when the kernel supports it, without the developer having to ask for it. A service written in Go and deployed on a vulnerable kernel may be accepting connections through a path its operations team did not know was open.

The kernel fix comes with a curious asymmetry. In Red Hat Enterprise Linux 8, 9 and 10, MPTCP is disabled by default, which reduces exposure although it does not eliminate the flaw. In Ubuntu the situation is different because support is more widespread. That difference explains why the same CVE has such disparate ratings.

## The severity discrepancy that no one explains

Red Hat, Amazon, Oracle and SUSE place the flaw between 5.9 and 8.1, while the kernel CVE team maintains a 9.8. The gap is not a whim. Each vendor scores according to its default configuration and its threat model, not according to the worst theoretical scenario.

A 5.9 suggests that an uncommon condition is needed to exploit it. A 9.8 indicates that anyone with network access can attempt it without credentials. Both figures can be correct at the same time and that is precisely the trap: teams that prioritize by number postpone patches that their own deployment does need.

For a systems administrator the useful reading is different. The question is not what score the CVE has, but whether your kernel compiles MPTCP, whether there are Go listeners and whether the service is exposed to the Internet. If you are not clear on where to start, [the guide to reading a CVE in ten minutes](https://observatorioblockchain.com/en/cybersecurity/como-leer-un-cve-si-te-afecta/) orders the same checks without jargon. Three specific checks that are worth more than any scale.

## How to know if you are exposed in three commands?

The first check is the kernel version and the stable branch. If the system runs 6.12.113, 6.18.56, 7.2.10 or 7.3-rc7 or later, the flaw is fixed. The affected versions go up to 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.112, 6.18.55 and 7.2.9 according to the inventory published by Hungary's national cybersecurity center.

The second is whether MPTCP is active. The system parameter `net.mptcp.enabled` shows the status at runtime, and the command `ss -M` lists the sockets that are using that stack. A system with MPTCP disabled is not immune, because the flaw lives in shared code, but it reduces the entry path that the exploit takes advantage of.

The third is to review which services are listening and in what language they are written. A binary compiled with Go 1.24 or higher on a vulnerable kernel is the combination that should be patched first, especially if it handles public traffic. The rest can wait for the usual maintenance window without drama, as long as there are network compensations in front.

## The patch without reboot and its limits

For environments where a reboot is not possible, the vendor TuxCare published a kernel patch without reboot for Debian 13 on October 29 and kept the versions for Debian 12 and Oracle Linux UEK under construction, as well as those for RHEL 8, 9 and 10, AlmaLinux, Rocky Linux and CloudLinux. It is a legitimate solution for databases and appliances, with the caveat that it only covers the kernel and not user libraries.

The coverage has another limit that is worth stating out loud. A live patch replaces functions in memory and cannot always touch structures that are already in use. In a use-after-free flaw like this one, the window between loading the module and full stability is measured in hours, not seconds. Those who prefer the slow path have [the Linux 7.4 merge window](https://observatorioblockchain.com/en/technology/linux-7-4-merge-window-6753-cambios/) as a reference for when the fixes arrive in each branch.

Teams that chose to wait for the official version from their distribution have a pending decision. In Ubuntu, kernels 5.15.0-203.213 for 22.04, 6.8.0-149.149 for 24.04, 7.0.0-43.43 for 26.04 and 7.3.0-10.10 for 26.10 were in preparation at the close of this week, with no publication date. Ubuntu 20.04 and earlier remain vulnerable and no longer receive new kernels through the normal channel.
