Critical Linux kernel flaw lets Anthropic's AI write a remote root exploit — Ubuntu and RHEL still haven't patched it

A flaw in the Linux kernel's TCP handshake code, tracked as CVE-2026-98432, remains unpatched on Ubuntu and Red Hat Enterprise Linux as of October 29, even though a fix has been available upstream since October 5 and several competing distributions closed the hole weeks ago, as first reported by The Hacker News. The Linux kernel's own CVE team rates the bug 9.8 out of 10 on the CVSS severity scale, though individual vendors including Red Hat, Amazon, Oracle, and SUSE score it lower, between 5.9 and 8.1.

The vulnerability matters because of both its severity and its age: the underlying bug was introduced in Linux 4.4, meaning it has sat in the kernel for roughly a decade before anyone found it. It is a race condition in the code that sets up new IPv4 TCP connections — when a connection completes, the kernel hands the IP options stored with the pending request to the new socket without copying them. If that socket is destroyed while another CPU is still reading those options to answer a resent connection request, the kernel ends up reading memory it has already freed, a bug class known as use-after-free. IPv6 connections aren't affected because the kernel copies the options rather than sharing a pointer to them. Triggering the flaw requires a server with a multipath TCP (MPTCP) listener, a feature that lets a single connection use several network paths at once and that an application must explicitly request when opening a socket.

Anthropic security researcher Xinyang Ge is credited in the kernel fix for reporting the bug, and the company said separately, in research published October 27, that its AI model went further and wrote a working exploit chain that uses the flaw to let a remote attacker with no credentials run commands as root on a vulnerable server. That detail turns an abstract memory-safety bug into a demonstrated path to full remote takeover, and it adds CVE-2026-98432 to a small but growing list of vulnerabilities where an AI model — not a human researcher — produced functioning attack code.

The patching gap is the more immediate operational problem. Debian, Oracle Linux, SUSE, and Amazon Linux have all shipped fixed kernel packages, with Amazon Linux 2023 patching as recently as October 24 and 29. Ubuntu lists its 22.04, 24.04, 26.04, and 26.10 releases — along with its cloud, FIPS, and real-time kernel variants — as "work in progress" with no fixed package released, and still rates its older 16.04, 18.04, and 20.04 releases as vulnerable with no patch planned timeline given. Red Hat Enterprise Linux 8, 9, and 10, along with OpenShift 4, list the fix as "deferred to a future update" with no release date. Neither Canonical nor Red Hat has published when a fix will ship, leaving organizations running MPTCP-enabled services on either distribution exposed to a credential-less root exploit that is already known to work.