Cybersecurity researchers have discovered a critical flaw in the Linux kernel that allows attackers to remotely crash servers without authentication or access to the system. The NoisyTCP vulnerability went undetected for approximately 11 years and threatens servers running Nginx, Apache, SSH, and other TCP-based services.
The vulnerability was discovered by security researcher Hyunwoo Kim, who specializes in identifying security flaws. NoisyTCP was assigned the identifier CVE-2026-99288 and is associated with a race condition in the Linux kernel’s TCP networking stack. To carry out an attack, an attacker only needs to send specially crafted network packets to an accessible server, even without an account or any system privileges.
According to the researcher’s technical report on GitHub, the flaw affects common TCP services, including Nginx, Apache, and SSH running with default configurations. Since these technologies are widely used to host websites, web applications, and provide remote administration, millions of servers worldwide could potentially be at risk.
The vulnerability was introduced into the Linux kernel back in 2015 and remained undetected for approximately 11 years. According to the researcher, it affects virtually all Linux distributions, although exploitability depends on each server’s network configuration.
NoisyTCP: Linux Kernel TCP Pre-auth Remote Corruption
CVE-2026-99288. "11 years" latent. Patch it.
The era of KVM escapes is fading, and the era of TCP remote exploits has begun.https://t.co/yA3pUapOPY
— V4bel (@v4bel) November 15, 2026
Exploiting NoisyTCP requires at least two incoming network traffic processing (RX) operations to run simultaneously. This can occur on servers with multiple packet reception queues or mechanisms that distribute network workloads across CPU cores.
Due to a flaw in TCP connection handling, the Linux kernel may access a network socket whose memory has already been freed and reused for another object that has not yet been fully initialized. This can trigger a critical kernel failure, known as a kernel panic, causing the entire server to crash.
What makes this vulnerability particularly dangerous is that the attack does not affect just a single application. If an attacker exploits the flaw against a web server running inside a container, the resulting crash can bring down the entire host system.
The researcher also tested how quickly the vulnerability could be reproduced. During tests involving Nginx, Apache, and SSH in QEMU/KVM virtual machines, the attack succeeded in approximately 5 seconds to 2 minutes. On a physical server running Nginx with eight RX queues, exploitation took around 14 hours, even with a specially configured exploit.
NoisyTCP can also potentially lead to a use-after-free condition, where software accesses memory that has already been released. Under certain circumstances, such vulnerabilities can create opportunities for arbitrary code execution. However, the researcher emphasizes that NoisyTCP is not currently a fully developed remote code execution (RCE) vulnerability. The confirmed threat is limited to Denial-of-Service (DoS) attacks capable of crashing servers.
A fix for NoisyTCP has already been merged into the mainline Linux kernel and some distribution packages. However, not all systems, including popular enterprise Linux distributions, will receive the update at the same time.
Red Hat rated CVE-2026-99288 at 7.5 on the CVSS scale and recommended that administrators temporarily disable SYN cookies or limit incoming traffic processing to a single RX queue. However, disabling SYN cookies weakens protection against SYN flood attacks, so installing the security patch remains the preferred solution.
This is the second serious vulnerability discovered in the Linux kernel’s TCP mechanisms in less than a month. Previously, Anthropic researchers reported CVE-2026-98432, which received a severity score of 9.8 and allows attackers to remotely execute arbitrary commands with root privileges without authentication.
The discovery of NoisyTCP demonstrates that even long-established and extensively tested Linux components can contain dangerous flaws for many years. Server administrators are advised to check for available kernel updates and install the relevant security patches as soon as they become available for their distributions.