Critical Linux Kernel TCP Flaw (CVE-2026-98432) Exploited by AI, Scores 9.8 CVSS
A critical vulnerability in the Linux kernel's TCP implementation, CVE-2026-98432, with a CVSS score of 9.8, is being exploited by an AI model to achieve remote code execution with root privileges.
A severe vulnerability, designated CVE-2026-98432, has been discovered in the Linux kernel's TCP implementation, earning a critical CVSS score of 9.8 out of 10. This flaw impacts widely used Linux distributions, including Ubuntu and Red Hat Enterprise Linux (RHEL), and has reportedly been weaponized by an AI model capable of generating exploits.
Anthropic disclosed on October 27 that its AI model successfully created an exploit for this vulnerability. The exploit allows a remote attacker, without any prior authentication, to execute arbitrary commands with root privileges on affected Linux servers, particularly those configured with a multipath TCP (MPTCP) listener. This capability significantly elevates the risk posed by the flaw, turning a potential kernel bug into an immediate threat.
As of October 29, Ubuntu and RHEL had not yet released patched kernel versions, leaving their users exposed. While other vendors like Red Hat, Amazon, Oracle, and SUSE have assigned scores ranging from 5.9 to 8.1, the Linux kernel's own CVE team maintains the highest severity rating of 9.8, underscoring the potential impact.
The vulnerability stems from a race condition within the code responsible for establishing new IPv4 TCP connections. Specifically, it's a use-after-free bug where the kernel mistakenly reads from memory that has already been deallocated. This type of memory corruption can lead to unpredictable behavior, including system crashes or arbitrary code execution.
The fix for this bug was credited to Anthropic researcher Xinyang Ge. While the official CVE record and the kernel patch itself do not explicitly mention Multipath TCP (MPTCP), Anthropic's analysis linked the exploitability to this feature. MPTCP allows a single connection to utilize multiple network paths simultaneously, and its implementation in Linux requires specific application requests when opening a socket.
Anthropic stated that the vulnerability was patched upstream and subsequently backported to all maintained stable kernel branches. However, the delivery of this fix to end-users is dependent on the respective Linux distribution vendors. Debian, Oracle Linux, SUSE, and Amazon Linux have begun shipping patched packages for some of their kernel versions. In contrast, Ubuntu and RHEL are still in the process of developing and releasing their fixes, with Red Hat indicating that patches are deferred to future updates and Ubuntu classifying the work as 'in progress' without a firm release date.
The discrepancy in severity scoring among vendors highlights the complexities of assessing such vulnerabilities. The kernel team's 9.8 score assumes an attacker can induce a race condition by sending specific TCP connection requests with IPv4 Record Route or Timestamp options. Red Hat, however, initially rated it as Moderate (5.9), citing the difficulty of winning the race and primarily predicting kernel crashes. This rating was later updated in a machine-readable advisory file to 'Important' with an 8.1 score, acknowledging higher confidentiality and integrity impacts, a sentiment echoed by Oracle and SUSE.
This critical flaw underscores the ongoing challenges in securing the Linux kernel, the backbone of much of the internet's infrastructure. The involvement of AI in exploit generation further accelerates the threat landscape, demanding rapid patching and robust security practices from both vendors and users to mitigate the risks associated with CVE-2026-98432.