CVE-2026-98432: Protect Your WordPress VPS and Bare-Metal Servers Before Ubuntu Patches

CVE-2026-98432 Linux kernel TCP flaw: cartoon server with a glowing red TCP cable and a broken padlock

This page has had its content updated on November 5, 2026 EST by Jordan

Content Error or Suggest an Edit

Notice a grammatical error or technical inaccuracy? Let us know; we will give you credit!

Found this useful? Managing WP is written in the gaps between client work.
If it saved you some time, a coffee keeps it going.

Introduction

⚡ Short on time? The key facts and fixes are in the TL;DR.

A critical bug in the Linux kernel’s TCP code, CVE-2026-98432, went viral on October 29. The kernel’s own CVE team rates it 9.8 out of 10, and Red Hat and Canonical now agree it’s Critical. It sits in the code that accepts new IPv4 TCP connections, so almost every internet-facing server has it, and it has been there since kernel 4.4 in early 2016.

Where things stand on November 5: Debian, Oracle Linux, Amazon Linux, SUSE and Red Hat have shipped fixed kernels, and so have Rocky Linux and AlmaLinux. Ubuntu, which GridPane, RunCloud and most VPS stacks run on, has its fixed kernels in the -proposed testing pocket for 22.04 and newer but hasn’t released them to normal updates yet. Researchers have also reported suspected scanning for vulnerable servers since November 2.

The one thing to do today: patch and reboot anything with a fixed kernel available. On Ubuntu, block the packets that trigger the bug and restart anything listening with MPTCP, then patch the moment Canonical releases the kernel.

Looking for Support?

Need help implementing this articles content? We can help!

TL;DR

  • What: a remote, no-login use-after-free in the Linux kernel’s IPv4 TCP handshake (CVE-2026-98432, CVSS 9.8). Kernels from 4.4 up to the fixed releases are affected.
  • Fixed: Debian, Oracle Linux, Amazon Linux, SUSE, RHEL 8/9/10 (November 1), Rocky and AlmaLinux. Install the kernel and reboot.
  • Ubuntu: fixed kernels for 22.04, 24.04 and 26.04 are in -proposed only; 26.10 is fixed; 20.04 and older have nothing yet.
  • Can’t patch yet: drop IPv4 TCP packets carrying IP options (one nftables rule), turn MPTCP off and restart anything that was listening with it, and lock web ports to Cloudflare.
  • MPTCP isn’t required: Red Hat and Canonical both say turning it off isn’t a fix. Go 1.24+ services (Caddy, exporters, agents) listen with MPTCP by default.
  • Status: no public exploit and no confirmed exploitation, but suspected scanning since November 2.

How It Works

It’s a use-after-free in the kernel’s IPv4 TCP connection setup. IPv4 option data attached to a half-open connection can be freed before an RCU grace period ends, so a race during the handshake can leave the kernel touching memory that’s already been released. Because this happens before any application authentication, an attacker needs no login and no user interaction, just the ability to reach a listening TCP port. IPv6 isn’t affected.

The headline exploit came from Anthropic’s Frontier Red Team: an unreleased Claude research model turned the bug into remote root against servers with a multipath TCP (MPTCP) listener, leaking kernel addresses in about 48 minutes and getting a root shell a few minutes later. Anthropic hasn’t published the exploit or its test configuration, and we won’t link anything that claims to be one.

What We Know So Far

  • Oct 15: CVE published; the kernel team assigned the 9.8 score on October 24. Debian shipped its fix on October 16.
  • Oct 27: Anthropic disclosed its AI-built exploit; coverage peaked on October 29.
  • Nov 1: Red Hat shipped fixed kernels for RHEL 8, 9 and 10 and raised its rating to Critical (9.8). Rocky Linux and AlmaLinux followed within days (RFF, ma.ttias.be).
  • Nov 2: Canonical published fixed kernels for 22.04 and newer to the -proposed testing pocket, with release to normal updates once testing finishes (Ubuntu).
  • Nov 2–3: Marcus Hutchins reported what his team believes is scanning for the bug: MPTCP connections to port 443 from residential proxies and VPNs. His own scan found about 2.1 million servers answering MPTCP on 443, which measures exposure, not vulnerable kernels (The Stack).
  • Nov 5: Ubuntu’s fixed kernels were still not in normal updates. No public exploit and no confirmed exploitation in the wild.

MPTCP is not a requirement. Only Anthropic’s write-up ties the exploit to MPTCP. Red Hat says MPTCP is not a required condition and that hosts with it disabled remain affected, and Canonical says security updates are the only appropriate mitigation. Treat turning MPTCP off as narrowing exposure, not as a fix.

Affected Versions and Exposure

The upstream fix is in stable kernels 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3. Anything from 4.4 up to those versions is affected. Distributions backport fixes without changing the version string, so check the kernel changelog rather than uname -r (command below).

  • RHEL: fixed in 4.18.0-553.182.1.el8_10 (8.10), 5.14.0-687.64.1.el9_8 (9.8) and 6.12.0-211.72.1.el10_2 (10.2). The RHEL 9 real-time kernel and OpenShift’s RHEL CoreOS were still listed as affected.
  • Ubuntu 22.04, 24.04, 26.04: fixed in 5.15.0-203.213, 6.8.0-149.149 and 7.0.0-43.43, currently in -proposed only. The 24.04 HWE 6.17 kernel has no fix staged yet.
  • Ubuntu 26.10: fixed in 7.3.0-10.10.
  • Ubuntu 16.04, 18.04, 20.04: fix pending (Ubuntu Pro/ESM). 14.04 is not affected.

Typical WordPress stacks (Nginx/OpenLiteSpeed, PHP-FPM, MySQL/MariaDB, Redis) don’t open MPTCP sockets, to our knowledge. But the kernel side is switched on (net.mptcp.enabled = 1) on GridPane servers on every Ubuntu version and on Proxmox VE hosts, and Go 1.24 and later listen with MPTCP by default whenever the kernel allows it. Any Go binary on a public port (Caddy, exporters, agents, small tools, a lot of Docker containers) is likely an MPTCP listener.

WordPress Hosting Providers and Control Panels

Unlike most server bugs, this one reaches straight into WordPress hosting: it’s in the kernel every Linux web server runs. On a managed platform the provider patches the kernel; on a VPS you manage with GridPane, RunCloud or a control panel, you do, and the panel can’t patch the kernel for you.

  • GridPane (Ubuntu): no public statement seen as of November 5. MPTCP is enabled on every Ubuntu version GridPane uses. Assume vulnerable until you’ve installed a fixed kernel and rebooted.
  • RunCloud (Ubuntu): no public statement seen as of November 5. Same advice.
  • cPanel/WHM: on AlmaLinux and CloudLinux, the fixed kernels are out; on Ubuntu, same as above.
  • SiteGround says it live-patched its servers the same day the flaw made headlines, with no reboots needed (SiteGround).
  • 1Panel now disables MPTCP by default as a mitigation, as of v2.4.2 (1Panel).
  • KernelCare / TuxCare had a rebootless patch for Debian 13 as of October 31 and no date yet for Ubuntu (TuxCare).
  • Managed WordPress hosts (Kinsta, WP Engine, Cloudways and similar): check the provider’s status page or ask support whether hosts have been patched or live-patched.

Infrastructure Beyond WordPress

Every Linux box that accepts TCP connections is in scope, so don’t stop at the web servers. Prioritise the ones that face the internet:

  • Proxmox VE hosts: Debian-based, so a fixed kernel is available; MPTCP is enabled by default.
  • Go services: Caddy, Prometheus exporters, monitoring agents, step-ca and similar tools listen with MPTCP by default on Go 1.24+.
  • Docker hosts and Kubernetes nodes: containers share the host kernel, so patch the host. On AWS, ECS on EC2 instances and EKS nodes need a kernel update and reboot or a fresh AMI (AWS).
  • Bastions, VPN endpoints, mail, database and backup servers with public ports.
  • Load balancers and appliances built on Linux: ask the vendor.

Immediate Mitigation Steps

1. Check whether the running kernel is fixed

Look in the kernel’s changelog for the CVE or the fix’s patch title. Some fixed Ubuntu kernels were built before the CVE number existed, so they only mention the patch title. Anything above 0 means the kernel you’re running has the fix.

# Debian / Ubuntu
zgrep -cE 'CVE-2026-98432|free inet_opt and ireq_opt' /usr/share/doc/linux-modules-$(uname -r)/changelog.Debian.gz

# RHEL / Rocky / AlmaLinux
rpm -q --changelog kernel-core-$(uname -r) | grep -cE 'CVE-2026-98432|free inet_opt and ireq_opt'

2. Patch and reboot wherever a fix exists

Debian, RHEL, Rocky, AlmaLinux, Oracle, Amazon Linux, SUSE and Ubuntu 26.10 hosts can patch now. See Installing / Applying the Fix below.

3. Block the trigger on unpatched hosts

The bug is triggered by IPv4 TCP connection requests that carry IP options, which normal traffic doesn’t use; RFC 7126 already recommends filtering them. This nftables rule drops IPv4 TCP packets with IP options before connection tracking. It was published and tested by RFF and Atlas, and works alongside ufw and firewalld.

nft add table ip cve_2026_98432
nft add chain ip cve_2026_98432 pre '{ type filter hook prerouting priority -300; }'
nft add rule ip cve_2026_98432 pre meta l4proto tcp ip hdrlength gt 5 counter drop

Three caveats: it doesn’t survive a reboot unless you add it to your nftables config, it also drops forwarded TCP packets with IP options if the server routes traffic, and don’t use it on CIPSO-labelled networks (SELinux or Smack NetLabel). Remove it once you’re on a fixed kernel.

4. Turn MPTCP off and restart its listeners

This only removes the condition Anthropic’s exploit used. Turning it off stops new MPTCP sockets but doesn’t close listeners that are already open, so restart whatever ss -Mlnp lists and run it again until it’s empty. For Go services you can also set GODEBUG=multipathtcp=0 in their environment.

sysctl net.mptcp.enabled
ss -Mlnp
sysctl -w net.mptcp.enabled=0
echo "net.mptcp.enabled = 0" > /etc/sysctl.d/99-disable-mptcp.conf
# then restart each service ss -Mlnp listed, and re-run: ss -Mlnp

Leave out -t when checking: ss -Mltnp also lists every ordinary TCP listener and makes everything look exposed.

5. Plan the Ubuntu reboots now

Line up maintenance windows so the fixed kernel goes out the day Canonical releases it. If a server can’t wait, see the -proposed option below.

Hardening Checklist

  • Firewall every port that doesn’t need to be public. Fewer listening sockets means less to reach.
  • Restrict SSH to known IPs or put it behind a VPN or Cloudflare Tunnel.
  • Lock web ports 80/443 to Cloudflare’s IP ranges on proxied sites so the origin isn’t directly reachable.
  • Inventory Go binaries and containers listening on public interfaces.
  • Leave net.ipv4.tcp_migrate_req at its default of 0; request socket migration opens a similar path.
  • Confirm unattended-upgrades is on and that someone actually reboots after kernel updates.

Installing / Applying the Fix

Debian

Fixed kernels are available. Update and reboot.

apt update && apt full-upgrade -y
reboot

RHEL / AlmaLinux / Rocky

Fixed since November 1 (Red Hat advisories RHSA-2026:93414, RHSA-2026:93412 and RHSA-2026:93413 for RHEL 8, 9 and 10). Rocky and AlmaLinux have the rebuilt kernels.

dnf update kernel
reboot

Ubuntu

Ubuntu 26.10 is fixed in normal updates. For 22.04, 24.04 and 26.04 the fixed kernels aren’t in normal updates yet; when they land, run the same apt update && apt full-upgrade and reboot. Ubuntu Pro users can run pro fix CVE-2026-98432 once it’s released. Livepatch had no notice for this CVE as of November 5.

Ubuntu: installing the kernel from -proposed (if you can’t wait)

-proposed is Ubuntu’s testing pocket. Canonical made it available for people who run their own kernel acceptance tests, while it finishes testing on its certified hardware. If a server is exposed and the block rule isn’t an option, you can install just the fixed kernel from it. Snapshot first, try it on one server, and keep a way back to the old kernel in GRUB. On GridPane- or RunCloud-managed servers, check with the provider before installing an unreleased kernel.

Canonical’s own commands enable -proposed for the main component and upgrade only your kernel meta-packages from it. Packages in -proposed are never picked up automatically, so nothing else upgrades from it.

sudo add-apt-repository --component main --pocket proposed http://archive.ubuntu.com/ubuntu
sudo apt update
dpkg-query -W -f '${source:Package}\t${binary:Package}\n' | awk '$1 ~ "^linux-meta" { print $2 }' | xargs sudo apt install -t $(. /etc/os-release && echo $VERSION_CODENAME)-proposed --only-upgrade
sudo reboot

After the reboot, run the changelog check from step 1. Once Canonical releases the kernel to normal updates, you can remove -proposed again (sudo add-apt-repository --remove --component main --pocket proposed http://archive.ubuntu.com/ubuntu).

Mitigations to Reduce Exposure

1. Drop IPv4 TCP packets with IP options (tested against the trigger, not the exploit)

Covered in step 3 above. It blocks the packets that trigger the bug; RFF confirmed the rule drops them, but nobody outside Anthropic has the exploit to test against.

2. Disable MPTCP (confirmed for Anthropic’s exploit path only)

Covered in step 4. It removes the one demonstrated path, but Red Hat and Canonical both say the bug doesn’t need MPTCP. Remember to restart existing MPTCP listeners.

3. Reduce reachable TCP listeners (unconfirmed)

The bug is in connection setup, so an attacker has to reach a listening port. Closing ports at the firewall or provider level shrinks the target.

4. Enable the Cloudflare WAF (limited)

The WAF works at HTTP, above where this bug lives, so it can’t filter the attack itself. What does help is proxying sites through Cloudflare and locking origin 80/443 to Cloudflare IPs, so attackers can’t open TCP connections to your web server directly. SSH and any other open ports are still reachable.

5. Take the Site Offline (doesn’t apply here)

Our usual last resort, GridPane HTTP Authentication or Suspend Site and RunCloud Site Authentication, works at the web server level. The kernel still accepts the TCP connection first, so it doesn’t protect you from this bug. If a host is genuinely at risk, firewall it or take it off the network instead.

Detecting Exploitation

1. Watch the kernel log for crashes

A failed race can leave traces. Look for use-after-free, general protection faults or oopses.

journalctl -k --since "2026-10-15" | grep -iE "use-after-free|general protection|BUG:|Oops"

2. Look for unexpected root processes and logins

ps -eo user,pid,ppid,lstart,cmd --sort=lstart | tail -n 30
last -a | head -n 30

3. Check for new accounts and SSH keys

awk -F: '$3 == 0 {print $1}' /etc/passwd
find / -name authorized_keys -newer /etc/hostname 2>/dev/null

4. Review outbound connections

A compromised host often calls home. Check for listeners and connections you don’t recognise.

ss -tunap | grep -v -E "127.0.0.1|::1"

If anything looks off, treat the host as compromised: isolate it, snapshot it, and rebuild rather than clean.

FAQ

Is this being exploited?

Not confirmed. There’s no public exploit and no report of the bug being used in attacks, but researchers have seen what they believe is internet-wide scanning for MPTCP listeners on port 443 since November 2. Treat that as a reason to move faster.

Is my server safe if MPTCP is off?

Safer, not safe. Turning MPTCP off blocks the exploit Anthropic described, but Red Hat and Canonical both say the bug doesn’t depend on it. And it only applies to new sockets: restart anything that was already listening with MPTCP.

Should I install Ubuntu’s -proposed kernel?

Only where you can’t wait and can’t use the block rule. It’s the real fix, but it hasn’t finished Canonical’s testing. For most servers, block the trigger now and take the kernel from normal updates when it lands.

Does Cloudflare protect me?

Only if your origin only accepts web traffic from Cloudflare. Cloudflare can’t filter a kernel TCP bug, but it can stop attackers reaching your origin’s web ports directly.

Do I need to reboot?

Yes. A kernel update only takes effect after a reboot, unless you’re using a live-patch service.

Found this useful? Managing WP is written in the gaps between client work.
If it saved you some time, a coffee keeps it going.

Updates

  • November 5, 2026: Red Hat shipped fixed RHEL 8, 9 and 10 kernels on November 1 (Rocky and AlmaLinux followed); Ubuntu’s fixed kernels for 22.04 and newer are in -proposed and 26.10 is fixed. Added reports of suspected scanning, Red Hat’s and Canonical’s position that MPTCP isn’t required, the IPv4-options drop rule, the restart caveat for existing MPTCP listeners and the Go setting, a kernel changelog check, and steps for installing the Ubuntu kernel from -proposed. Added a TL;DR and sections on hosting providers and wider infrastructure.
  • October 31, 2026: Added that GridPane servers (every Ubuntu version) and Proxmox VE hosts have MPTCP enabled. Corrected the MPTCP listener check to ss -Mlnp; the earlier ss -Mltnp also listed ordinary TCP listeners.
  • October 31, 2026: First version. Ubuntu and RHEL still have no fixed kernel; Debian, Oracle and Amazon Linux have shipped fixes.

Sources

0 Shares:

You May Also Like