CVE-2026-98432 – When a TCP handshake leads all the way to root
The patch situation has improved markedly: almost all major distributions have now released patched kernels — Ubuntu (22.04, 24.04, 26.04, 26.10), Debian 12 and 13, RHEL 8/9/10 plus extended-support streams, AlmaLinux, Rocky, Oracle Linux, Fedora, Arch, SLES/Leap 16 and Amazon Linux. Concrete package versions and remaining gaps (Ubuntu HWE, Ubuntu 20.04/18.04 without Pro, openSUSE Leap 15.6) are listed in the affected systems and fixes section. This site itself has been running Ubuntu 24.04 with the patched kernel 6.8.0-149-generic since November 6.
The essentials first
A security flaw in TCP connection setup is especially inconvenient for server operators: the faulty code runs inside the kernel, i.e. below web server, application and login. That is exactly where CVE-2026-98432 sits. Red Hat warns of possible code execution by remote attackers without credentials and rates the flaw as critical.
Source: Red Hat Security Bulletin RHSB-2026-016.
Publicly reachable servers belong at the top of the review. What matters is a verifiably patched kernel package and its activation. An updated CMS or container image does not answer this question.
The figures also deserve a closer look: Red Hat rates this CVSS 3.1: 9.8 with low attack complexity. Amazon Linux rates it 8.1 with high attack complexity. The difference lies in AC:L versus AC:H; this does not allow a blanket all-clear for your own installation.
Ratings: Red Hat CVE entry and Amazon Linux CVE entry.

How the bug occurs
When a new TCP socket is created, it inherits a pointer to IPv4 options. In parallel, another CPU core can process a retransmitted SYN and read the same options for the SYN-ACK response. If the new socket is destroyed in the meantime, its memory may already be freed: the other code path accesses an invalidated region – a use-after-free.
The fix delays the release via kfree_rcu() until already running RCU readers are finished. Affected are the free paths in net/ipv4/af_inet.c and net/ipv4/tcp_ipv4.c. The kernel notification also describes a related path in the migration of connection requests.
IPv6 needs a differentiated view: the first flaw does not affect this path because options are copied there. For migration with CALIPSO, however, the notification names a similar problem that requires a separate fix. Claiming "IPv6 is completely safe" would therefore be too far-reaching.
MPTCP, Go and the RCE proof
On October 27, 2026, Anthropic published a research report with a concrete example for this CVE. According to the team, a remote attacker was able to execute commands as root on a Linux server with a Multipath-TCP listener – without login and user interaction. The report first describes defeating kernel address randomization and then taking over control flow.
For the described proof of concept the team cites 48 minutes to the information leak and another five minutes to command execution. These times apply to the research team's attempt; they are not a general prediction for other systems. Interestingly, according to the report, lost races can only make the handshake fail. A missing crash is therefore not a reliable security indicator.
Source: Anthropic, section 5: Zero-click remote code execution in Linux.
A notice by the Go team archived on oss-security explains: since Go 1.24, Linux listeners use MPTCP by default. GODEBUG=multipathtcp=0 can disable this default at runtime. However, explicit settings via net.ListenConfig.SetMultipathTCP and net.Dialer.SetMultipathTCP take precedence.
The Go team also stresses: the flaw is not limited to MPTCP. Systems without MPTCP listeners could be vulnerable through similar attacks. Disabling MPTCP is therefore a possible interim measure for certain services, not a replacement for the kernel fix.
Source: Go security notice, archived on oss-security on Oct 30, 2026.
On the threat situation: an exploit proof described by the research team is different from a documented attack campaign. The reports evaluated here provide no basis for claiming widespread active exploitation. Equally, the absence of such a report does not prove that no attacks are happening.
Affected systems and fixes
The kernel CVE team traces the introduction back to Linux 4.4 and lists the following fixed upstream versions. The list is not a substitute for the package proof of your own distributor.
| Branch | Fixed in |
|---|---|
| 5.10 | 5.10.272 |
| 5.15 | 5.15.223 |
| 6.1 | 6.1.190 |
| 6.6 | 6.6.159 |
| 6.12 | 6.12.113 |
| 6.18 | 6.18.56 |
| 7.2 | 7.2.10 |
| 7.3 | 7.3 |
Source: Linux kernel CVE team: affected and fixed versions.
Debian: patches are out
Debian is covered: Bookworm (linux 6.1.190-1, DLA-4841-1, plus linux-6.12 6.12.113-1~deb12u1) and Trixie (linux 6.12.113-1, DSA-6568-1); sid and the upcoming forky are fixed too. Only the combined look at security sources and the actually installed binary package is trustworthy: older packages in the regular archive may continue to appear vulnerable alongside it.
Source: Debian Security Tracker.
Ubuntu: generic kernels are out, HWE branches still open
The generic kernels have now been released: Ubuntu 22.04 (linux 5.15.0-203.213), 24.04 (linux 6.8.0-149.149) and 26.04 (linux 7.0.0-43.43) — each since November 5 to 6 — and 26.10 (linux 7.3.0-10.10, in the security pocket since November 2). The cloud and specialist kernels (AWS, GCP, Azure, Oracle, IBM, NVIDIA, OEM, low-latency, GKE) have followed by November 6. Still open: the generic HWE branches of 22.04 and 24.04 (status “work in progress”) and Ubuntu 20.04, 18.04 and 16.04, for which no fix is planned without an Ubuntu Pro subscription. Notably, as of November 6 there is neither an Ubuntu Security Notice nor Livepatch coverage for this CVE.
Source: Ubuntu CVE status.
RHEL, AlmaLinux, Rocky and Oracle Linux
Red Hat released kernels for RHEL 8.10, 9.8 and 10.2 on November 1 (RHSA-2026:93414, RHSA-2026:93412, RHSA-2026:93413); the extended-support streams (8.4–8.8, 9.2–9.6, 10.0) followed by November 4. AlmaLinux (8.10, 9.8, 10.2), Rocky Linux and Oracle Linux RHCK (8.10, 9.8) are fixed as well; Oracle's UEK kernels have been since October 16.
Fedora, Arch and beyond
Fedora (43–45) and Arch (including linux-lts and linux-hardened) were already safe by mid-October on the basis of the upstream fixes. Amazon Linux 2 and 2023 published all fixed kernels from October 24 to 30. Proxmox VE ships the fix from proxmox-kernel-7.0.14-25-pve (PSA-2026-00080-1). SLES 16.0/Leap 16.0 are covered by 6.12.0-160000.40.1; SLES 15 SP7, SP4–SP6 (LTSS) and 12 SP5 (LTSS) have replacement kernels or live patches. Only the openSUSE Leap 15.6 replacement kernel was still unconfirmed as of this update.
Package statuses: Red Hat bulletin and Amazon Linux package overview.

Check, patch, verify
For operations I recommend a short, documented chain: identify system → assign vendor fix → install update → boot patched kernel → check services. A single version number is not enough for that.
# On the affected Linux host: running kernel and distribution
uname -r
cat /etc/os-release
# TCP listeners and associated processes (with admin rights)
sudo ss -lntp
# MPTCP and migration configuration, if supported by the kernel
sysctl net.mptcp.enabled net.ipv4.tcp_migrate_req
# Debian/Ubuntu: installed kernel packages with full version
dpkg-query -W -f='${Package} ${Version}\n' 'linux-image*'
# RPM systems: installed kernel packages
rpm -q kernel kernel-core
These queries are an inventory, not an exploit test. An enabled MPTCP switch does not prove a reachable MPTCP listener. Conversely, a disabled switch does not certify that the kernel fixes the underlying CVE.
Cross-check the full package version against the matching vendor advisory. With a backport, an older-looking version string can already be fixed. Then install the released security update through the existing package management and perform the required reboot in the maintenance window. Only accept a livepatch as evidence if its vendor explicitly covers this specific CVE for the running kernel.
Containers and virtual machines
In an ordinary Docker or LXC environment, the host kernel belongs in the review; an image refresh does not replace it. For a VM, the guest kernel must be considered separately. For a container host this means: check the kernel outside the containers and, after the reboot, also verify the services running on it.
- Map the new output of
uname -rto the patched package or vendor proof. - Test reachability and function of the relevant services after the reboot.
- With multiple nodes, tick off each host individually; a patched node does not prove the state of the cluster.
- Only remove temporary measures deliberately after the fix is confirmed active.
What helps until the update
Red Hat recommends, as a temporary risk reduction, filtering incoming IPv4 TCP packets with IP options. The bulletin explicitly states that this does not provide complete protection. IP options sit in the IPv4 header; ordinary TCP options are distinct from this.
The concrete nftables configuration including testing and rollback is in the vendor bulletin. It should be tested against your own network structure. An INPUT rule shown there is in particular not blanket proof for all forwarded container traffic.
My operational recommendation: close unnecessarily public TCP services and, where practical, limit necessary access to known networks or VPN. For Go services, the MPTCP switch described above can additionally help. Document every interim measure with an owner and a rollback criterion so it does not displace the actual patch.
My assessment
The combination of a flaw in the network path and a described remote-root proof makes this CVE an urgent operational topic. For prioritization I would first look at publicly reachable hosts and known MPTCP services, then systematically work through the remaining affected kernels.
The most reliable sign-off reads: vendor fix verified, patched kernel active, services checked. That is more telling than a green update dialog or a supposedly non-critical version number.
Sources and verifiability
Research and editorial date: November 6, 2026. Technical statements and package statuses refer to primary sources. Vendor pages may change after this date. Operational recommendations are phrased as such; the illustrations are AI-generated symbolic images.
- Linux kernel CVE team – cause, affected files and fix versions
- Anthropic – research report from Oct 27, 2026, section 5
- Go team – security notice, archived on oss-security
- Red Hat RHSB-2026-016 – risk and interim measure
- Red Hat CVE entry – CVSS rating
- Debian Security Tracker – fixed source packages
- Ubuntu – status per kernel package
- Amazon Linux – rating and package advisories
- Continuously maintained overview: which Linux kernels are still vulnerable (ma.ttias.be)