Four Linux Exploits Published: Local Root via Bugs 10–21 Years Old

Security researcher Asim Manizada published four working proof-of-concept exploits for Linux kernel vulnerabilities on September 18, 2026, each enabling local privilege escalation to root. The bugs — dubbed DirtyAH6, TUNderflow, PPPoEject, and DiagSpill — have lurked in the codebase for 10 to 21 years and were discovered through an AI-assisted research process. Patches are available in upstream kernel versions from 5.10.270 to 7.2.4, but the release of functional PoCs drastically lowers the threshold for weaponization.

DirtyAH6: Out-of-Bounds Write in IPv6 Header

The first flaw, CVE-2026-80844, resides in the kernel's IPsec subsystem. The function ipv6_rearrange_rthdr() fails to verify that the segments_left field matches the addresses actually present in the IPv6 routing header. This mismatch allows writing past the allocated buffer's boundaries.

According to The Hacker News, Manizada assessed remote exploitation as extremely limited. The researcher's direct quote is clear: "looks extremely difficult". The remote path requires narrow conditions — an IPv6 router with AH transport mode configured — and even then reduces to a crash denial-of-service, not remote root. Root escalation remains a local path.

TUNderflow and PPPoEject: Two Faces of Memory Corruption

CVE-2026-81000, dubbed TUNderflow, exploits an integer underflow in the TUN/TAP driver. The receive-headroom value is reused as an allocation size, but when Open vSwitch supplies an oversized value, the size_t calculation wraps below zero and the kernel allocates an incorrectly sized buffer. Memory corruption follows immediately.

CVE-2026-68121, PPPoEject, is a use-after-free in the PPPoE subsystem. The kernel retains a pointer across a call to dev_hard_header(), but this function can free and reallocate the buffer transparently to the caller. The reference pointer becomes dangling, enabling access to already-freed memory.

Both require unprivileged user namespaces to reach full exploit conditions. The feature is enabled by default in many modern Linux distributions, including containerized environments and shared hosting.

DiagSpill: The Dangerous Exception That Needs No User Namespaces

CVE-2026-74469 is the most severe flaw by potential impact. Its CVSS 8.8 score — per metrics reported by SecurityOnline.info — reflects a broader attack surface: the vector is network, with low privileges and no user interaction required.

The mechanism is an integer overflow in the SCTP diag subsystem. A 16-bit counter wraps at 65,536, causing sctp_diag to reserve zero bytes while copying the entire association list. The result is a massive heap overflow of roughly 8 MiB. According to the dossier, this bug is reachable without unprivileged user namespaces, eliminating an entire mitigation layer that protects the other three flaws.

Manizada reported that neither AppArmor nor SELinux blocked the exploits in testing. The quote, from SecurityOnline.info, is verbatim: "AppArmor and SELinux did not block the exploits in testing".

"this batch likely ends the public phase of his AI-assisted bug hunting" — Asim Manizada, quoted by The Hacker News

The AI Signature in the Patch Commit

A distinctive detail emerges from the attribution chain. The fix commit for DirtyAH6 contains an 'Assisted-by' line explicitly crediting the artificial intelligence tooling used in the research. This is not a footnote: it is the first documented trace in the Linux kernel development flow where the discovery process is formally recognized in the fix attribution.

Manizada's methodology, detailed in The Hacker News, maps kernel memory management and applies structured reasoning on object layout — an approach the researcher himself indicates concludes his public phase of AI-assisted bug hunting.

Patching as a Version Chase: Kernels and Distributions

Patches are available upstream in versions 5.10.270, 5.15.221, 6.1.188, 6.6.157, 6.12.109, 6.18.50, and 7.2.4, per the summary table published by SecurityOnline.info. Specific commit hashes are identified for each flaw.

The sources cite only upstream kernel versions. No information emerges on which Linux distributions have already released patched packages in their repositories: the lag between upstream release and distro channel availability is an unquantified variable in the dossier. Systems tracking vanilla kernels with rapid update cadences are covered; long-term support installations with conservative backport cycles accumulate exposure.

What to Do Now

• Verify the running kernel version and compare it against patched releases: 5.10.270, 5.15.221, 6.1.188, 6.6.157, 6.12.109, 6.18.50, 7.2.4.
• Disable unprivileged user namespaces where not strictly necessary, mitigating the attack path for DirtyAH6, TUNderflow, and PPPoEject.
• Monitor distribution repositories for updated kernel packages: upstream patch availability does not imply immediate distro channel availability.
• Prioritize patching on multi-user servers, shared hosting, container environments, and cloud instances where an attacker with limited access can attempt root escalation.

Why the Disclosure Timeline Matters

Manizada notified the vulnerabilities to security@kernel.org in mid-July 2026. The roughly two-month interval between responsible disclosure and PoC publication aligns with kernel development cycles, but creates a measured exposure window: systems not updated before September 18 now face public, verified exploits — not theoretical attacks.

The risk is not theoretical. The combination of working PoCs, a bug with extended attack surface (DiagSpill without user namespaces), and failure of major mandatory access control systems to block the exploits indicates that remaining barriers lie entirely in patching, not runtime mitigation.

Frequently Asked Questions

Is there confirmation of in-the-wild exploitation?

No. The dossier documents no observed real-world attacks: the current threat stems from public PoC availability, which reduces the time and expertise needed for weaponization.

Why does CVE-2026-74469 have a higher CVSS than the others?

The attack vector is network (AV:N) versus local (AV:L) for the other three, with identical attack complexity and CIA impact. This vector difference shifts the score from 7.8 to 8.8.

Does AI-assisted discovery change the expected volume of vulnerabilities?

Manizada himself indicates this release concludes the public phase of his AI-assisted work. The data does not support generalization to other researchers or volume trends, but it documents that AI tooling produced concrete results on decade-old codepaths not previously analyzed systematically.

Sources

• https://thehackernews.com/2026/09/public-exploits-released-for-four-linux.html
• https://securityonline.info/linux-kernel-lpe-quartet-disclosed/
• https://blog.netmanageit.com/public-exploits-released-for-four-linux-kernel-flaws-that-enable-local-root/
• https://thehackernews.com/
• https://thehackernews.com/p/upcoming-hacker-news-webinars.html
• https://thehackernews.com/search/label/Threat%20Intelligence
• https://thehackernews.com/search/label/Vulnerability
• https://thehackernews.com/search/label/Cyber%20Attack