Critical Linux Kernel Flaw Enables Remote Root Attacks on MPTCP Servers
Security researchers at Anthropic develop an exploit for a critical Linux kernel vulnerability, tracked as CVE-2026-98432, that lets unauthenticated attackers remotely execute commands as root on servers running Multipath TCP (MPTCP) listeners. The flaw earns a severity score of 9.8 out of 10 from the Linux kernel CVE team, and it also affects ordinary TCP connections, though remote code execution is not yet demonstrated on servers without an MPTCP listener. MPTCP is a TCP feature that allows a single connection to use multiple network paths simultaneously to improve performance and reliability.
The vulnerability is a use-after-free flaw in the kernel's IPv4 TCP networking code that stems from a race condition in how the kernel handles TCP connection requests. It affects kernels dating back to version 4.4, including widely used distributions such as Ubuntu, Red Hat Enterprise Linux, and Rocky Linux. A similar issue can also occur during TCP request socket migration when specific kernel options or BPF programs are enabled. Although triggering the flaw requires no authentication, Red Hat classifies it as a potential denial-of-service issue with a lower severity score of 5.9, arguing that exploiting the race condition is more complex than simply sending a malicious network packet.
Anthropic researchers demonstrate that the vulnerability can nonetheless be exploited for remote code execution under specific conditions, publishing an exploit developed with the help of an unreleased Claude AI model. Administrators running affected distributions are urged to apply kernel updates as they become available, particularly on servers that expose MPTCP listeners to untrusted networks. The incident highlights growing concerns about race-condition flaws in core networking code, where timing-based exploitation can turn theoretical weaknesses into practical, remote root-level attacks.