Executive Summary
In September 2026, security researcher Asim Manizada disclosed a 'quartet' of Linux kernel vulnerabilities identified through AI-assisted memory mapping. These flaws—DirtyAH6, TUNderflow, PPPoEject, and DiagSpill—stem from memory safety issues in networking subsystems that have existed for 10 to 21 years. While primarily local privilege escalation (LPE) threats, two of the flaws (DirtyAH6 and DiagSpill) can be triggered remotely to cause system crashes (DoS) or, in highly specific lab conditions, remote code execution.
The attack chain typically involves exploiting unprivileged user namespaces to gain the necessary network capabilities to trigger out-of-bounds writes or use-after-free conditions. DiagSpill is particularly notable as it requires no special privileges or namespaces, only the availability of the SCTP module. Exploits for all four flaws have been made public, increasing the risk for multi-user systems and container environments where attackers might attempt container escapes.
Impact is high for organizations running legacy or unpatched Linux kernels across cloud, server, and embedded environments. While there are no reports of active exploitation in the wild, the availability of functional PoCs tuned for common distributions necessitates immediate patching to stable kernel versions including 5.10.270, 6.1.188, 6.6.157, and 7.2.4 or later.
Key Details
Threat Name
AI-Assisted Linux Kernel LPE Quartet
Affects
Linux kernel IPsec AH6 (IPv6), Linux kernel TUN/TAP virtual network devices, Open vSwitch, Linux kernel PPPoE, Linux kernel SCTP (sctp_diag), Linux kernel 2.6.12 through 7.2.2, Linux kernel 4.6 through 7.2.3, Linux kernel 2.6.12 through 7.1.5, Linux kernel 4.7 through 7.1.7
Adversary
—
Malware/Tools
DirtyAH6, TUNderflow, PPPoEject, DiagSpill
