# Anthropic Says Unreleased Claude Built Root Exploit for Linux Kernel Flaw

- URL: https://usa-24.com/cve-2026-98432-linux-kernel-tcp-flaw/
- Publisher: USA 24
- Section: Cybersecurity
- Published: 2026-10-30T08:49:39.296Z
- Updated: 2026-10-30T08:49:39.307Z
- Topics: Linux Security, Kernel Vulnerabilities, Ubuntu, Red Hat, AI Security

> Anthropic says an unreleased Claude model built a working root exploit for a critical Linux kernel TCP flaw, and Ubuntu and Red Hat still lack a fixed kernel.

## Key takeaways

- Anthropic says an unreleased Claude model built a working root exploit for CVE-2026-98432, a CVSS 9.8 Linux kernel TCP flaw, and shared the finding on October 27, 2026.
- As of October 30, Ubuntu had not released a fixed kernel, and Red Hat listed its fix for RHEL 8, 9 and 10 as deferred with no date.
- Debian, Oracle Linux and SUSE have shipped fixed kernels, and Amazon Linux 2023 has shipped kernel 6.12 for its optional kernels.
- Every kernel from Linux 4.4 onward is affected until a fixed stable release is installed, and upstream fixes shipped October 11 and October 15.
- No in-the-wild exploitation has been reported, and the flaw was not in CISA's Known Exploited Vulnerabilities catalog as of October 29.

An unreleased Claude model built a working root exploit for a critical Linux kernel bug, Anthropic says, and two of the Linux distributions that run much of the US server fleet still lack a fixed kernel. The flaw, tracked as CVE-2026-98432 and scored 9.8 on CVSS, can let a remote attacker run commands as root with no login and no user action. Anthropic's Frontier Red Team shared the finding on October 27. As of October 30, the gap between upstream fixes and enterprise packages was still open.

The upstream fix is not the problem. Stable kernels with the patch shipped on October 11 and October 15, according to [Atlas](https://atlasum.net/blog/cve-2026-98432-linux-ipv4-tcp-uaf). Debian, Oracle Linux and SUSE have shipped fixed kernels. Ubuntu has not, and Red Hat has deferred its fix for RHEL 8, 9 and 10 with no date.

## What Anthropic says its model did

Anthropic's research post, as summarized by [Cybersecurity News](https://cybersecuritynews.com/linux-kernel-flaw-root-commands/), describes a lab test rather than a real-world incident. After a first prompt from a researcher, Claude wrote an exploit that targets Linux servers with a multipath TCP (MPTCP) listener. Anthropic describes the root cause as a race condition that leads to a use-after-free in the kernel networking stack.

The trigger is a TCP connection request, known as a SYN packet, that carries an IPv4 Record Route or Timestamp option. Third-party analysis from [GetRFF](https://getrff.com/blog/cve-2026-98432-linux-tcp-ip-options-block/) reports that the exploit needed about 48 minutes to get past KASLR, the kernel's address randomization, and about five more minutes to get a shell. That puts the total near 53 minutes. Anthropic has not published the exploit, so those timings cannot be checked independently.

No direct quotations from Anthropic appear in this story. The account rests on secondary coverage of the research post, and the full write-up was not available for verbatim quotation. Treat the timings as reported figures, not verified results.

## Which servers are exposed, and why the MPTCP detail matters

The Linux kernel CVE team published the record on October 15, 2026. The fix was committed on October 6, according to [Rescana](https://www.rescana.com/post/linux-kernel-tcp-cve-2026-98432-use-after-free-prepare-now). Every kernel from Linux 4.4 onward is affected until a fixed stable release is installed, per [PatchMon](https://patchmon.net/blog/cve-2026-98432-linux-kernel-tcp-use-after-free-patch).

The demonstration used MPTCP listeners, and that detail creates a trap. Administrators who do not run MPTCP may assume they are safe. Anthropic has not said whether servers without MPTCP can be rooted, and the bug is an IPv4 TCP handshake issue rather than a multipath-only one. Until a vendor or Anthropic says otherwise, treat any reachable TCP listener on an affected kernel as potentially exposed.

Rocky Linux and AlmaLinux rebuild from RHEL, so they carry the same Red Hat gap. Shops running those distributions should not expect a fix before Red Hat ships one.

## Patch status for CVE-2026-98432 by distribution

| Distribution | Status as of October 30, 2026 |
|---|---|
| Ubuntu | No fixed kernel released; the tracker lists fixes as in progress |
| Red Hat Enterprise Linux 8, 9, 10 and OpenShift 4 | Fix listed as deferred with no date; Red Hat says a fix is not guaranteed |
| Rocky Linux and AlmaLinux | Rebuild from RHEL, so they face the same unpatched exposure |
| Debian | Fixed kernel shipped |
| Oracle Linux | Fixed kernel shipped |
| SUSE | Fixed kernel shipped |
| Amazon Linux 2023 | Shipped kernel 6.12 for optional kernels |

Upstream stable releases with the fix include 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3. The upstream commit, e3f33b0a1d89, was authored by Eric Dumazet and credits Anthropic researcher Xinyang Ge as the reporter, according to [Penligent](https://www.penligent.ai/hackinglabs/cve-2026-98432/). Distribution kernels often backport fixes without changing the upstream version string, so administrators should confirm status with their vendor rather than compare version numbers alone.

## Why the severity scores disagree

The kernel CNA and Ubuntu rate the flaw 9.8 under CVSS 3.1. Red Hat's rating varies across coverage. [Panelica](https://panelica.com/blog/cve-2026-98432-why-this-linux-kernel-bug-is-a-98-on-ubuntu-but-a-59-on-red-hat) frames the gap as a 9.8 on Ubuntu but a 5.9 on Red Hat, while another source lists Red Hat's rating as 8.1 Important. The available sources do not explain the difference, so treat the discrepancy as unresolved.

Teams that triage by vendor severity could be tempted to push this down the list. That would be a mistake. A lower Red Hat score does not change the fact that the fix is missing, and Anthropic says a working exploit already exists in its lab.

## What to do this week

- Inventory every Linux host and record its distribution and kernel version.
- Flag hosts that run TCP listeners reachable from the internet, then rank internal listeners reachable from less trusted network segments.
- Check each vendor's CVE page for CVE-2026-98432, starting with Ubuntu's tracker and Red Hat's entry.
- Install the fixed kernel as soon as your distribution ships it, then reboot. The new kernel does not run until the machine restarts.
- Test any temporary workaround before deploying it. A third-party nftables rule blocks the trigger until vendor kernels ship, but vendors have not verified it. [Cyber Kendra](https://www.cyberkendra.com/2026/10/cve-2026-98432-linux-kernel-tcp-flaw-ubuntu-rhel.html) reports that Red Hat says no available mitigation meets its standards for ease of deployment and stability.
- Monitor for unusual spikes in repeated SYN packets aimed at a single listener.

## What to watch next

Ubuntu's CVE page is the most useful signal to follow. A change from work in progress to released would close the most visible gap in the current picture. A dated Red Hat fix would give administrators a planning window, and a Known Exploited Vulnerabilities listing from CISA would start a federal clock. CISA had not added the CVE to its catalog as of October 29.

The pattern is familiar. CISA added Fragnesia, tracked as CVE-2026-46300, to its catalog on October 28, with a federal deadline of October 31 and a requirement for forensic triage, according to [HOL](https://hol.org/guard/security/campaigns/cve-2026-46300-fragnesia-linux-kernel-cisa-kev-root). Fragnesia was disclosed May 13, 2026, and its patch shipped the same day, but existing Dirty Frag patches did not cover it, per [Tenable](https://www.tenable.com/blog/fragnesia-cve-2026-46300-faq-about-new-linux-kernel-xfrm-esp-in-tcp-priv-esc). This is the second privilege-escalation bug in the Linux kernel to land in a short window, and administrators are patching both at once.

The Ubuntu gap also resembles the one covered in [Node2Shell's Brotli flaw story](https://usa-24.com/node2shell-vulnerability-nodejs-brotli/), where a KEV listing arrived while patches lagged. Federal deadlines are landing fast elsewhere too, as the [Cisco firewall VPN zero-day](https://usa-24.com/cve-2026-76418-cisco-firewall-patch/) shows with its November 1 date.

The next concrete test is whether Anthropic publishes its exploit. If it does, defenders will learn quickly whether servers without MPTCP are in reach, and that answer will shape how many administrators can wait for their vendor's kernel.

## Frequently asked questions

### Is my Linux server vulnerable to CVE-2026-98432?

Every kernel from Linux 4.4 onward is affected until a fixed stable release is installed. Anthropic has not said whether servers without MPTCP can be rooted, so administrators should check any reachable TCP listener on an unpatched kernel. Confirm your distribution's status on its CVE page before deciding what to do.

### Has Ubuntu or Red Hat released a fix?

As of October 30, 2026, Ubuntu had not released a fixed kernel, and its tracker lists fixes as in progress. Red Hat lists the fix for RHEL 8, 9 and 10 and OpenShift 4 as deferred with no date and says a fix is not guaranteed. Debian, Oracle Linux and SUSE have shipped fixed kernels.

### Did Anthropic's model attack real systems?

Anthropic's research post, as summarized in secondary coverage, describes the exploit as a lab test rather than a real-world incident. No in-the-wild exploitation has been reported for this CVE. Anthropic has not published the exploit, so its reliability outside the lab is unknown.

### Why do sources give different severity scores for this flaw?

The kernel CNA and Ubuntu rate the flaw 9.8 under CVSS 3.1. Red Hat's rating appears as 5.9 Moderate in some sources and 8.1 Important in another. The available sources do not explain the gap, so teams should not lower their priority based on the lower figure while the fix is missing.

### Is there a temporary workaround?

A third-party guide describes an nftables rule that blocks the trigger until vendor kernels ship. Vendors have not verified that rule, and Red Hat says no available mitigation meets its standards for ease of deployment and stability. Test any workaround in a staging environment before deploying it to production servers.

## Sources

1. [Claude AI Exploits Critical Linux Kernel Flaw to Execute Root Commands Without User Interaction — Cybersecurity News](https://cybersecuritynews.com/linux-kernel-flaw-root-commands/)
2. [Linux's October 2026 TCP Use-After-Free: CVE-2026-98432 — Atlas](https://atlasum.net/blog/cve-2026-98432-linux-ipv4-tcp-uaf)
3. [Linux Kernel CVE-2026-98432: Critical TCP Use-After-Free, No RHEL or Ubuntu Fix Yet — Rescana](https://www.rescana.com/post/linux-kernel-tcp-cve-2026-98432-use-after-free-prepare-now)
4. [CVE-2026-98432 Kernel Flaw: Ubuntu, RHEL, Debian Status — PatchMon](https://patchmon.net/blog/cve-2026-98432-linux-kernel-tcp-use-after-free-patch)
5. [CVE-2026-98432: how to block the Linux TCP bug until Ubuntu and Red Hat ship a kernel — GetRFF](https://getrff.com/blog/cve-2026-98432-linux-tcp-ip-options-block/)
6. [Critical Linux Kernel TCP Flaw CVE-2026-98432 Still Unpatched on Ubuntu and RHEL — Cyber Kendra](https://www.cyberkendra.com/2026/10/cve-2026-98432-linux-kernel-tcp-flaw-ubuntu-rhel.html)
7. [CVE-2026-98432: Linux TCP Use-After-Free Risk and Patch Validation — Penligent](https://www.penligent.ai/hackinglabs/cve-2026-98432/)
8. [CVE-2026-98432: Why This Linux Kernel Bug Is a 9.8 on Ubuntu but a 5.9 on Red Hat — Panelica](https://panelica.com/blog/cve-2026-98432-why-this-linux-kernel-bug-is-a-98-on-ubuntu-but-a-59-on-red-hat)
9. [Fragnesia (CVE-2026-46300) — HOL](https://hol.org/guard/security/campaigns/cve-2026-46300-fragnesia-linux-kernel-cisa-kev-root)
10. [CVE-2026-46300 (Fragnesia): Linux Kernel ESP-in-TCP LPE FAQ — Tenable](https://www.tenable.com/blog/fragnesia-cve-2026-46300-faq-about-new-linux-kernel-xfrm-esp-in-tcp-priv-esc)