---
title: Critical Vulnerability Rated 9.8 Exploited by Anthropic's AI
url: https://www.elseif.net/critical-vulnerability-rated-98-exploited-by-anthropics-ai
published: 2026-11-01T17:03:11+00:00
language: en
section: Models
source: https://ascii.jp/elem/000/004/447/4447103/?rss
organizations: anthropic, cve-2026-98432, linux
publisher: elseif
---

# Critical Vulnerability Rated 9.8 Exploited by Anthropic's AI

Anthropic announced on October 27 that it demonstrated an attack exploiting the Linux kernel vulnerability "CVE-2026-98432," allowing remote command execution with administrator privileges. The vulnerability has been rated as severe, scoring 9.8 out of a maximum of 10.

The issue lies within the IPv4 communication processing of the Linux kernel. It is a defect known as "use-after-free (UAF)," which refers to referencing memory that has already been freed during the handling of TCP connections. This vulnerability can be exploited over a network without the need for authentication or user interaction.

In its demonstration, Anthropic targeted a "MPTCP" listening server utilizing multiple communication paths. It was reported that the AI-generated attack code enabled remote execution of commands with root privileges without authentication.

The CVE team, responsible for managing Linux kernel vulnerability information, publicly announced the issue on October 15, and a fix has been incorporated into the stable version. However, according to updates from Ubuntu on October 29, multiple versions are still undergoing remediation or are considered vulnerable.

Server administrators are advised to check the patch status provided by their respective Linux distribution and apply the necessary updates to mitigate this risk.