Linux kernel security patches: 9.8-rated TCP flaw gets fixes, AI finds more bugs
Red Hat has shipped patches for a Linux kernel TCP remote code execution flaw rated 9.8, and Ubuntu has released fixes for testing, according to coverage on 3 November. The bug, tracked as CVE-2026-98432, is a use-after-free in the kernel's IPv4 TCP handling.

Red Hat has shipped patches for a Linux kernel TCP flaw rated 9.8 out of 10. Ubuntu has released fixes for testing. The updates were reported on 3 November.
The bug is tracked as CVE-2026-98432. It is a use-after-free in the kernel's IPv4 TCP code, according to vulnerability write-ups circulating since late October. This is a remote code execution flaw in the networking stack. Almost every Linux server, container host and cloud instance runs that stack. A 9.8 score on the CVSS scale means an attacker needs no account and no local access. Administrators who manage internet-facing machines should treat the patch window as short. It is unclear how many distributions have a stable, fully tested build out. It is also unknown whether exploit code is in wide circulation.
What the flaw does and why the score is high
The vulnerability sits in the kernel's handling of IPv4 TCP connections. A use-after-free means the kernel frees a piece of memory and then keeps using it. An attacker can steer that toward running their own code. Coverage from 28 October flagged that the flaw affects Ubuntu, Red Hat and other distributions. Two days later, on 30 October, a CVE entry for CVE-2026-98432 appeared. The 9.8 rating is the network-facing maximum short of a perfect 10. It requires no privileges and no user interaction, and it hits confidentiality, integrity and availability.
The timeline of responses has been uneven. On 2 November, reports said attackers had started scanning in the wild for the kernel RCE, and Ubuntu released fixes. The same day, an exploit for the flaw was reported to have been produced by an AI model. That detail matters. It suggests the gap between a patch landing and working attack code appearing is now measured in hours, not weeks.
"It looks to me like our security processes need to invert somewhat, since just one person searching for the issue class ... is sufficient to alert someone else's agent and let them get exploit code. This is wild."
That quote comes from Anil Madhavapeddy, professor of computer science at Cambridge and a core maintainer of the OCaml compiler, in an article covered by InfoQ on 5 October. Madhavapeddy described fixing a path-traversal bug. He saw probes in his webserver logs with the exact bug pattern minutes after he opened the pull request. InfoQ also cited a study in which a GPT-4 agent exploited 87% of vulnerabilities in a 15-vulnerability benchmark when given CVE descriptions, against 7% without them.
The patch pipeline is under new pressure
The Linux kernel's security process was already strained before this month. Debian's latest kernel security update carried 1,313 fixes, The Register reported on 5 October. GIGAZINE's coverage of the same update noted that tracking individual vulnerabilities one by one may no longer be workable as AI-assisted bug hunting scales up. That is the context in which a 9.8 network flaw lands: not a quiet month, but a flood.
Maintainers of other large open source projects describe the same shift. Nick Craig-Wood, creator of the rclone project, said in a Hacker News thread cited by InfoQ that the project received about 20 security disclosures through GitHub in its first 10 years, and more than 40 in the last month. Adrian Mouat of Chainguard, also quoted by InfoQ, argued that simply opening a pull request to fix an issue now puts users at risk. Attackers can build exploits before an updated release exists. His suggested fix, publishing releases before the associated source code, breaks a basic open source norm.
Postgres offers a counterexample to the idea that AI is driving every hard decision. Tomas Vondra, a Postgres contributor, wrote on 5 October that several major features were reverted from Postgres 19 and the release will slip about a month. Blog posts, including one by Elizabeth Christensen, had suggested AI-found bugs drove the reverts. Vondra went through the mailing list threads and concluded most of the issues were not found by AI. The revert followed a human review by Andres and too many post-commit problems found late.
Vondra's post is worth reading closely because it separates two things that get blurred. One is AI finding bugs faster. The other is a project choosing caution when a fix is too invasive late in a cycle. PostgreSQL has no official AI contribution policy yet, though one is expected soon. Its August 2026 patch for Postgres 18 carried 28 CVEs, according to Christensen's post as quoted by Vondra.
Disclosure is getting harder to control
The kernel community has already had a taste of accidental disclosure this cycle. On 21 October, several Linux kernel vulnerabilities were posted to the public kernel mailing list by mistake. Coverage the next day flagged that two of them were critical RCE bugs. A public mailing list is exactly the wrong place for an embargoed fix. Anyone watching, human or agent, can pick it up.
That incident sits alongside the Madhavapeddy argument: embargoes assume secrecy protects users, and agents can now work from partial clues. His three suggested mitigations are private vulnerability discussions, faster continuous releases, and rapid protocol-level mitigations before full patches are ready. None of them is simple for a kernel that ships through distributions on their own schedules.
The commercial tooling around kernel review is moving in the same direction. Sashiko is an agentic code review system for Linux kernel changes. It reviews commits from a local git checkout or from mailing lists at lore.kernel.org, using kernel-specific prompts and a multi-stage verification protocol. Its documentation notes that it sends patch data and surrounding files to the configured LLM provider. That raises its own question about where vulnerability details end up.
OpenClaw's audit gives a sense of what AI-assisted review finds in practice. The company said on 5 October that Trail of Bits submitted 27 private repository advisories and 3 hardening pull requests through OpenAI's Patch the Planet initiative. Of 24 severity-rated reports, 23 were confirmed vulnerabilities: 0 Critical, 2 High, 16 Medium and 6 Low. Trail of Bits used Codex-assisted workflows to search for issues, then checked findings manually. The most common problem was permissions being lost between steps. Follow-on work ran without the limits of the original request.
What administrators should do now
The practical advice for the TCP flaw is unglamorous. Check whether your distribution has published a fixed kernel package. If it has, install it and reboot. Ubuntu released fixes on 2 November and moved them to testing; Red Hat's patches were reported on 3 November. Hosts that cannot be patched immediately should be watched for unusual TCP behaviour. Scanning alone will not stop an exploit that reaches a vulnerable stack.
Two things are still unknown. Neither the number of machines exposed nor the number already compromised has been published. And public reporting has not described the exploit reportedly generated by an AI model in detail, so its reliability is unclear.
Longer term, the argument Madhavapeddy makes is that patching speed has to rise to match the speed of discovery. The kernel's current model was built for a slower attacker. Fixes go to distributions, and users patch on their own timetable. That assumption is now doing less work than it used to.
Sources
10- 01AI Agents Are Disrupting Open Source Security DisclosureEN
- 02Postgres: Are we reverting patches because of bugs found by AI?EN
- 03OpenClaw Completes Security Audit Through OpenAI's Patch the Planet InitiativeEN
- 04Sashiko: Agentic review of Linux kernel codeEN
- 05Linux containers now run natively on Windows via WSLEN
- 06Windows Developers Get a Built-In Linux Container Runtime: WSL ContainersEN
- 07Stop Killing Games movement sounds alarm on EU Kids ActEN
- 08Kagi Ends Orion Browser Development for Linux, Will Open-Source ItEN
- 09Altera Quartus Linux jtagd bug fixesEN
- 10Netkit: Specializing Linux Packet Delivery for Container NetworksEN
All figures and quotations in this text come from the sources listed below.
Content prepared by the editorial team with AI assistance.
Comments
0- No comments yet — be the first.