Linux kernel exploit enables remote root attacks on MPTCP servers

Security researchers at Anthropic demonstrated an exploit for a critical Linux kernel vulnerability that permits unauthenticated attackers to remotely execute commands as root on servers running Multipath TCP listeners. Tracked as CVE-2026-98432, the flaw is a use-after-free defect in the kernel IPv4 networking code affecting releases dating back to version 4.4. The Linux kernel CVE team assigned the vulnerability a severity rating of 9.8 out of 10, whereas Red Hat evaluated it at 5.9 due to exploitation complexity. As detailed in BleepingComputer's report, researchers used an unreleased Claude AI model to bypass memory protections and attain root access on an affected system in approximately 53 minutes.

This report covers the mechanics of the networking flaw, affected distributions, software exposure vectors, and immediate mitigations for Linux systems. It is intended for systems administrators, DevOps engineers, and security teams managing Linux servers with public network interfaces. Because newer programming language runtimes can automatically enable Multipath TCP on standard network sockets, services may expose vulnerable endpoints without administrator knowledge. Understanding which environments face exposure, how the underlying race condition operates, and what packet filtering rules block exploitation allows technical teams to defend infrastructure while awaiting distribution vendor patches.

What broke in Linux kernel exploit enables remote root

The issue originates from a use-after-free vulnerability inside the Linux kernel IPv4 networking code during TCP connection setup. Upstream kernel documentation indicates that a race condition occurs during inbound request handling, allowing retransmitted packets to access memory structures concurrently across separate processor cores. This desynchronization permits double reads of packet option lengths and invalid pointer reuse in kernel space. The Linux kernel CVE team assigned CVE-2026-98432 a critical severity score of 9.8 out of 10 because remote network clients can trigger the flaw across unauthenticated network boundaries without needing existing credentials.

The flaw affects Linux kernels dating back to version 4.4, meaning years of historical releases contain the problematic socket handling routine. While kernel maintainers have patched upstream branches, downstream distribution repositories must package and distribute the changes. The underlying memory flaw exists in standard IPv4 TCP handling, but demonstrated remote code execution currently targets Multipath TCP configurations. This operational nuance led Red Hat to classify the vulnerability as a potential denial-of-service condition with a 5.9 severity rating, before Anthropic proved that coordinated socket manipulation can achieve full remote command execution.

Who is exposed by Linux kernel exploit enables remote root

Multiple major enterprise Linux distributions remain vulnerable while maintainers prepare security packages. Canonical reports that Ubuntu 22.04 LTS, Ubuntu 24.04 LTS, Ubuntu 26.04 LTS, and Ubuntu 26.10 are affected, with kernel fixes actively in progress. Red Hat lists Red Hat Enterprise Linux 8, 9, and 10 as vulnerable, with fixes scheduled for upcoming updates, though legacy RHEL 7 is unaffected because its kernel lacks the vulnerable code. Support provider CIQ confirmed that Rocky Linux 8, 9, and 10 are similarly exposed. Systems running macOS remain completely safe, as Apple platforms rely on the XNU kernel rather than Linux networking code.

System exposure is amplified because common programming runtimes can activate Multipath TCP listeners without administrator knowledge. Starting in Go 1.24, the runtime enables MPTCP by default for TCP listeners on Linux whenever host kernels support the capability. Consequently, web services and proxies compiled with Go 1.24 or later may expose MPTCP endpoints unintentionally across public interfaces. Even though Multipath TCP is not universally enabled by default across general Linux software, its automatic activation in Go application stacks means standard network services can expose the vulnerable kernel path to remote attackers.

What to do now about Linux kernel exploit enables remote root

Administrators should verify kernel versions and apply distribution patches as soon as vendor updates become available. Debian has published patched kernel package 6.1.190-1 for Debian 12 Bookworm and version 6.12.113-1 for Debian 13 Trixie, alongside updates for Debian testing and unstable branches. Other enterprise Linux vendors, including Amazon Linux, Oracle Linux, and SUSE, have also delivered fixed packages for specific kernel versions. Systems teams should inspect their package managers, apply available updates immediately, and restart host machines to ensure running operating systems load the patched kernel binaries.

On hosts where distribution patches are not yet available, administrators should apply temporary network mitigations. Rocky Linux support provider CIQ advises that disabling MPTCP alone is insufficient, as ordinary TCP traffic can still trigger the underlying memory corruption. CIQ recommends temporarily blocking incoming IPv4 TCP packets that contain IP options before they reach vulnerable kernel code, using nftables firewall rules. Restricting public network access to trusted IP ranges or keeping vulnerable servers behind secure network gateways provides an additional layer of protection against unauthorized inbound connection requests until updated kernels are deployed.

How the Linux kernel exploit enables remote root issue works

The underlying vulnerability occurs when the kernel transfers ownership of socket options during connection processing. According to the advisory, "tcp_v4_syn_recv_sock() transfers ownership of ireq->ireq_opt to the child socket (newinet->inet_opt) without copying it." Concurrently, another CPU core processing a retransmitted SYN packet for the same request socket sends a SYNACK from tcp_check_req(). The functions tcp_v4_send_synack() and inet_csk_route_req() read ireq->ireq_opt under rcu_read_lock() only, after which ip_build_and_send_pkt() and ip_options_build() read opt->optlen twice, producing a race condition. A similar issue occurs during socket migration with net.ipv4.tcp_migrate_req or BPF_SK_REUSEPORT_SELECT_OR_MIGRATE.

Anthropic researchers demonstrated weaponization of this race condition using an unreleased Claude artificial intelligence model. The generated exploit bypassed kernel address space layout randomization defenses on a target server hosting an MPTCP listener, taking approximately 53 minutes to achieve arbitrary command execution as root. By sustaining carefully timed network interactions, the exploit manipulated kernel heap structures to escape memory boundaries and control execution flow. This demonstrated that despite the technical complexity cited by distribution vendors, the race condition provides a viable path to full remote host compromise under specific listener configurations.

What is still unknown about Linux kernel exploit enables remote root

While researchers achieved remote root access on servers running MPTCP listeners, it remains unknown whether remote code execution is feasible against standard TCP listeners without MPTCP active. Although the use-after-free defect resides in core IPv4 TCP code, researchers have not publicly demonstrated an exploit achieving command execution without an MPTCP listener present. Red Hat continues to evaluate the flaw as a denial-of-service crash vector due to the timing precision required. Whether security researchers or attackers can craft reliable exploit primitives on ordinary TCP sockets without MPTCP remains an unanswered technical question.

Uncertainty also surrounds the exact release timeline for security updates across remaining enterprise Linux distributions. Canonical has not yet published finalized kernel packages for affected Ubuntu releases, and Red Hat continues developing fixes for RHEL 8, 9, and 10 systems. Furthermore, while there are no confirmed reports of CVE-2026-98432 being exploited in the wild, defensive teams cannot predict how quickly independent actors might replicate the AI-generated exploit methodology. Organizations must therefore monitor distribution advisories closely and maintain temporary network filtering until verified vendor packages arrive across production server fleets.