CVE-2026-98432: Prove the Fixed Kernel Is Running, Not Just Installed

Published on November 16, 2026

A kernel update can finish successfully while the vulnerable kernel continues handling traffic. That is the operational trap behind the response to CVE-2026-98432: package installation is a milestone, not closure. For Linux administrators, the useful question is not simply whether automation ran, but whether every affected machine has crossed into a verified, patched running state.

AWS’s security bulletin, published October 30 and updated November 12, 2026, describes a memory issue in the Linux kernel’s IPv4 TCP handshake path affecting kernel versions 4.4 and later. Under specific conditions, crafted packets can cause a crash, memory disclosure or corruption. Researchers demonstrated unauthenticated remote code execution against hosts with a Multipath TCP listener. That specific research result must not be rewritten as proof that every exposed Linux listener is remotely exploitable.

The same bulletin says AWS had no evidence the issue had been leveraged at the time of its update. A serious vulnerability and a demonstrated exploit are not a confirmed breach of your fleet. They are, however, sufficient reasons to act promptly without waiting for an incident.

Build an evidence sheet for each host

Record four things together: the distribution and package source, the installed kernel package, the running kernel, and the vendor advisory establishing which package contains the fix. Add an owner and a deadline. A version string alone can mislead because distributions backport security fixes while retaining older upstream version numbers.

Read-only observations such as uname -r identify the running kernel. Your distribution’s package inventory identifies what is installed. Neither establishes remediation alone; compare both with the current vendor advisory. Keep the evidence in the maintenance ticket rather than trusting a green deployment badge.

Also record which hosts terminate external TCP connections, which are behind a terminating proxy, and which permit direct backend access. A diagram that says “behind a load balancer” is not enough. The specific listener type and reachable bypass paths matter. AWS distinguishes service-specific protections in its bulletin; do not extrapolate those statements to arbitrary proxies or another cloud.

Reboot is part of the change

AWS explicitly directs Amazon Linux customers to update the kernel and reboot, and describes an update plus reboot as complete remediation for this issue. For another distribution, follow that vendor’s current instructions rather than copying Amazon Linux package commands.

Schedule the restart with the same care as the package change. Check capacity, drain traffic where supported, preserve console access and verify application health before advancing to the next node. For a single VPS, tell stakeholders about the interruption and confirm that backups and recovery access work. Do not improvise firewall changes that could remove your only administrative connection.

After restart, capture the running kernel again and reconcile it with the approved package. Check that the expected application services returned, that external health checks pass and that no node silently booted an older fallback kernel. A successful SSH login is not an application acceptance test.

Fix tomorrow’s replacement machines too

A patched host built from a stale image can be replaced by another vulnerable host during autoscaling. Update the source image or provisioning definition through the normal reviewed process, then inspect a newly created replacement. For container workloads, identify the host kernel: rebuilding an application container is not equivalent to replacing the kernel underneath it.

Managed control planes and self-managed workers are different responsibilities. AWS states that customers remain responsible for self-managed EC2 instances and gives separate instructions for its container services. Translate the bulletin into a list of assets you own, rather than treating a provider’s infrastructure update as proof that all customer workloads are repaired.

Retain a rollback plan, but recognise its security implications. Booting the old vulnerable kernel may restore availability while reopening exposure. If rollback becomes necessary, involve the incident or service owner, apply vendor-supported exposure reduction where appropriate, and track the host as an unresolved exception.

Watch for trouble without mistaking silence for safety

Review unexpected service failures, unusual logins, process activity and outbound connections during the change. These observations can reveal operational failures or suspicious behavior, but a quiet dashboard does not prove a kernel vulnerability was never exploited. A compromised kernel can also undermine host-level visibility.

One experiment in this area is Unhaxor, an AI security engineer concept for Linux behavior monitoring and suspicious-activity alerts. Its public onboarding is staged early access. Its stated role is observation and investigation, not automatic blocking, patching or remediation; detection of this CVE has not been independently verified here. Editorial disclosure: LinuxBoxAdmin and Unhaxor share ownership. Evaluate it as a possible supplementary signal, not as a reason to postpone a restart.

AI-assisted research may accelerate discovery and analysis of future flaws, while also increasing the number of reports administrators must assess. That is a planning risk, not evidence that AI caused this vulnerability or any particular intrusion. A repeatable patch-and-proof workflow remains useful regardless of how a flaw was found.

Closure checklist

  • Every affected host has an owner and vendor-backed fixed-package reference.
  • Installed and running kernel evidence agree after the required restart.
  • Replacement images and newly launched nodes are checked, not just existing instances.
  • Service health and administrative recovery paths work.
  • Any postponed reboot or rollback remains an explicit, time-bounded exception.
  • Suspicious activity is investigated separately; patch completion does not erase an earlier compromise.

Close the ticket when the running fleet matches the evidence, not when the package manager stops printing output.