On this page (7 sections)
CVE-2026-98432 is a use-after-free in the Linux kernel's IPv4 TCP connection setup code. The kernel's own CVE team scores it 9.8 out of 10, and it can be triggered over the network without logging in. Fixed kernels have shipped for Debian, Oracle Linux, SUSE, Amazon Linux and, as of early November, Red Hat Enterprise Linux. Ubuntu's fixes were still in an opt-in testing channel at last report.
If you run Linux servers that listen on TCP ports, including web, mail, database and container hosts, this one is worth checking today. Vendors disagree on how severe it is, so the practical answer is to confirm your kernel version and schedule the update and reboot.
What happened
The flaw was published as a CVE on October 15, 2026. According to The Hacker News, kernel developers applied the upstream fix on October 5, and the kernel CVE team added the 9.8 score on October 24. The fix credits Anthropic researcher Xinyang Ge with the report.
Attention rose on October 27, when Anthropic said an unreleased internal AI research model wrote a remote root exploit for the bug. Per the same report, that exploit works against Linux servers with a multipath TCP (MPTCP) listener, and Anthropic has not published the code. Proof-only code that triggers the race and detects freed memory echoed in a SYN-ACK has appeared on GitHub, but it does not achieve code execution.
Red Hat shipped fixed kernels for RHEL 8, 9 and 10 on November 1, according to The Hacker News follow-up. That report says Canonical put fixes for Ubuntu 22.04, 24.04 and 26.04 into -proposed on November 2, with no release date for the regular channel.
Technical details
This is a race condition leading to a use-after-free. When a server finishes a TCP handshake, the function tcp_v4_syn_recv_sock() hands the IP options stored with the pending connection request to the new child socket without copying them. If the client retransmits its SYN, another CPU can still be building a SYN-ACK reply that reads those options while the child socket is destroyed and its options are freed with a plain kfree(). The result is a read of freed memory in ip_options_build(), as described in the GitHub advisory text.
A similar problem exists with request socket migration, which applies when net.ipv4.tcp_migrate_req=1 or a BPF_SK_REUSEPORT_SELECT_OR_MIGRATE program is in use. The upstream fix switches the freeing to kfree_rcu() in inet_sock_destruct() and tcp_v4_reqsk_destructor(), so memory is released only after other CPUs finish reading it.
IPv6 is not affected by the main issue, because the IPv6 path duplicates the options. The advisory notes a related migration issue in tcp_v6_reqsk_destructor() that is only reachable with CALIPSO and will be handled in a separate patch.
Why the severity scores differ
The kernel CVE team's vector is network, low complexity, no privileges, no user interaction, with high impact on confidentiality, integrity and availability. Its reasoning, as reported, is that both sides of the race are attacker-induced, using IPv4 Record Route or Timestamp options on connection requests. Red Hat's original 5.9 reflected that the race is hard to win, with a crash as the most likely outcome. Amazon Linux's 8.1 says the main outcome is a crash with possible memory corruption. Red Hat has since moved to 9.8. Treat the spread as disagreement about exploit reliability, not about whether to patch.
Who is affected
Any Linux system running an affected kernel and accepting IPv4 TCP connections is in scope, since the trigger is a connection request to a listening port. Vendor status as reported on October 29 and November 3:
| Distribution | Status | Fixed kernel or advisory |
|---|---|---|
| RHEL 8 / 9 / 10 | Fixed Nov 1, reboot required | 4.18.0-553.182.1.el8_10 (RHSA-2026:93414); 5.14.0-687.64.1.el9_8 (RHSA-2026:93412); 6.12.0-211.72.1.el10_2 (RHSA-2026:93413) |
| RHEL 7 | Not affected | Not needed |
| Ubuntu 22.04 / 24.04 / 26.04 | Fix in -proposed only | 5.15.0-203.213; 6.8.0-149.149; 7.0.0-43.43 |
| Ubuntu 20.04 / 18.04 / 16.04 | No fix yet | Older releases need Expanded Security Maintenance via Ubuntu Pro |
| Debian trixie / 12 | Fixed in October | linux 6.12.113-1 (DSA-6568-1); DLA-4841-1 and DLA-4850-1 for 12 |
| Oracle Linux 7-10 | Fixed Oct 16 | ELSA-2026-500397, -500398, -500399 |
| SUSE | Fixed for Micro 6.2 and Leap 16.0; others in testing or in progress | kernel-default 6.12.0-160000.40.1 or later |
| Amazon Linux 2023 | Fixed for kernel6.18 and kernel6.12 | ALAS2023-2026-3251, ALAS2023-2026-3335 |
Ubuntu cloud, FIPS and real-time kernels are listed as work in progress, and Canonical's post covers only the generic kernel. Containers share the host kernel, so patching the host is what matters.
How to check your exposure
First find the running kernel. Compare the output with the fixed versions in the table above and your vendor's advisory. Run this on each server:
uname -rOn RHEL-family systems, this lists installed kernel packages so you can see whether the fixed build is installed, though you still must reboot into it:
rpm -q kernelOn Debian and Ubuntu, this shows the installed kernel image packages:
dpkg -l 'linux-image*' | grep '^ii'If you want to know whether MPTCP is in use on your network, the SANS ISC diary on MPTCP explains that MPTCP appears as TCP option 30 and shows how to spot it with Wireshark, Suricata or Zeek. Remember that the bug is in the plain IPv4 TCP path, so not seeing MPTCP does not mean you are safe.
What to do now
- Inventory Linux hosts that accept IPv4 TCP connections and record each running kernel with uname -r.
- On RHEL 8, 9 and 10, install the kernel from RHSA-2026:93414, RHSA-2026:93412 or RHSA-2026:93413, then reboot.
- On Debian, Oracle Linux, SUSE and Amazon Linux, install your vendor's fixed kernel package and reboot.
- On Ubuntu 22.04, 24.04 and 26.04, decide whether to use the -proposed kernel now or wait for the regular channel; Canonical has given no date.
- For Ubuntu 20.04, 18.04 and 16.04, check Ubuntu Pro Expanded Security Maintenance, since no fix is listed yet.
- Where you cannot patch quickly, consider the Red Hat nftables mitigation below, after testing it for breakage.
- After rebooting, run uname -r again and confirm the fixed version is actually running.
On RHEL-family systems, the standard update command pulls the fixed kernel once your repositories carry it:
sudo dnf update kernelOn Debian, refresh package lists and upgrade. Your meta-package pulls in the new kernel image:
sudo apt update && sudo apt upgradeUbuntu: the -proposed channel
Canonical's -proposed pocket is an opt-in testing channel and packages there do not install automatically. The Hacker News reports this procedure for administrators who need the fix early. It enables -proposed, upgrades the kernel meta-packages from it, and reboots. Use it on a test host first:
sudo add-apt-repository --component main --pocket proposed https://archive.ubuntu.com/ubuntu
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 rebootInterim mitigation from Red Hat
Red Hat published an nftables rule on October 30 that drops inbound IPv4 TCP packets carrying IP options, which blocks the known trigger. Red Hat says it cannot fully prevent exploitation and may break workloads that depend on IPv4 options. A CIQ kernel team member noted on oss-security that traffic between containers on the same host can bypass it. Load it only on the host being protected, and treat it as a stopgap:
table inet cve_2026_98432 {
chain input {
type filter hook input priority -310; policy accept;
ip protocol tcp ip hdrlength > 5 counter drop
}
}Detection and monitoring
No vendor has reported attacks, so there is no confirmed indicator to hunt for. The reported trigger is TCP connection requests carrying IPv4 Record Route or Timestamp options, followed by retransmitted SYNs. Legitimate internet traffic rarely uses these options, so network sensors that can log TCP header length or options may be useful. The same SANS diary shows a tcpdump header-length match and a Zeek event for option logging that you can adapt. Watch for unexpected kernel crashes or reboots on internet-facing hosts, which Red Hat and Amazon consider the most likely failure mode.
Also keep an eye on status changes: Ubuntu's regular-channel release, SUSE and Amazon Linux entries still marked pending, and any addition to CISA's Known Exploited Vulnerabilities catalog. If you run public websites on these hosts, certificate and header-change monitoring can help you notice unexpected changes, and the free SSL checker is a quick sanity check after maintenance reboots. Neither detects this kernel bug, so kernel patching remains the fix.
Timeline
| Date (2026) | Event |
|---|---|
| Oct 5 | Upstream kernel fix applied |
| Oct 15 | CVE-2026-98432 published |
| Oct 16-22 | Oracle Linux and Debian fixes released |
| Oct 24 | Kernel CVE team adds 9.8 score; Amazon Linux kernel6.18 fix |
| Oct 27 | Anthropic describes an AI-written exploit for MPTCP listeners; code not published |
| Oct 30 | Red Hat publishes nftables mitigation |
| Nov 1 | Red Hat ships RHEL 8, 9 and 10 fixes |
| Nov 2 | Canonical puts Ubuntu 22.04, 24.04 and 26.04 fixes in -proposed |
Sources
This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.
