Kernel maintainers have since released fixes for all four flaws, meaning systems running fully updated kernels are not affected.
Manizada published his technical analysis and working exploits on September 18 after coordinating with Linux distributions to give developers time to release patches. There are currently no reports of the vulnerabilities being exploited in real-world attacks. The published exploits were developed for specific kernel builds and can crash systems, making them primarily suited for isolated testing environments.
Despite those limitations, publicly available exploit code increases the risk for systems that have not been patched. Local privilege escalation vulnerabilities are particularly relevant on shared or multi-user systems, where an attacker who has already obtained limited access can potentially use the flaws to gain complete control.
Three of the vulnerabilities require unprivileged user namespaces to be enabled.
This Linux feature allows ordinary users to obtain root-like privileges inside an isolated environment and is enabled by default on many distributions. DirtyAH6, tracked as CVE-2026-80844, affects the IPv6 IPsec Authentication Header code. TUNderflow, CVE-2026-81000, affects TUN/TAP virtual network devices, while PPPoEject, CVE-2026-68121, targets PPP over Ethernet code.
DiagSpill, tracked as CVE-2026-74469, differs from the other three because it does not require user namespaces or special privileges.
Instead, it requires the SCTP networking module to be available.
Two vulnerabilities, DirtyAH6 and DiagSpill, can also be triggered remotely in limited circumstances, although the demonstrated remote impact is primarily system crashes. Manizada achieved remote root exploitation with DirtyAH6 in a controlled laboratory environment after first manipulating the target’s memory. He described achieving the same result remotely without that preparation as extremely difficult. He found no path to remote root with DiagSpill.
All four vulnerabilities are memory-safety flaws affecting different areas of Linux networking code. DirtyAH6 involves an out-of-bounds write in IPv6 IPsec handling, TUNderflow results from an integer wraparound in virtual networking code, PPPoEject is a use-after-free vulnerability, and DiagSpill involves a counter overflow that can result in a large out-of-bounds memory write.
The researcher said the flaws were discovered using an AI-assisted process designed to map kernel memory handling and reason about memory layouts.
The Linux fix for DirtyAH6 credits his custom AI tooling in its commit record. Manizada also previously disclosed another Linux kernel privilege-escalation flaw, OVSwrap, in July.
Administrators should update to a kernel containing all four fixes. The first stable Linux kernel releases containing the complete set are 5.10.270, 5.15.221, 6.1.188, 6.6.157, 6.12.109, 6.18.50 and 7.2.4. Distribution kernels use their own versioning, however, so users should check security advisories from their Linux distributor to confirm the fixes have been included.
If immediate patching is not possible, disabling unprivileged user namespaces can reduce exposure to DirtyAH6, TUNderflow and PPPoEject. Administrators can also disable AH6, TUN/TAP, PPPoE or SCTP features when they are not required. Manizada recommends patching rather than relying on feature restrictions because alternative exploitation paths may exist.
