"looks extremely difficult," Manizada wrote. That was his assessment of reaching remote root from afar using one of four newly published Linux kernel exploits — but the code is public now, and the practical risk depends on whether systems have received the recent fixes.
What was released and why it matters
On September 18, security researcher Asim Manizada published a technical write-up and working exploit code for four Linux kernel vulnerabilities that let a local user gain root privileges. The release followed a coordinated hold with Linux distributions so fixes could be issued first. Kernel maintainers have already fixed all four flaws in recent stable releases, so a system running an up-to-date kernel that includes those fixes is not affected.
That said, Manizada's exploit code is public, tuned to specific kernel builds, and can crash a machine; it is therefore intended for isolated test systems. There are, as of the publication, no reports that the four flaws have been used in real-world attacks. Nonetheless, public exploits increase the risk on shared systems where an attacker already has some access — for example, a low-privileged account on a multi-user server — and wants to escalate to full control.
The four flaws: DirtyAH6, TUNderflow, PPPoEject, DiagSpill
- DirtyAH6 — a bug in the IPsec code that handles the IPv6 Authentication Header. A routing-header field was trusted without verifying it against the number of addresses present, allowing a crafted packet to move an internal pointer out of bounds and write past a buffer.
- TUNderflow — in the TUN/TAP virtual network devices. A single value was used both as spare space and as a size; an oversized value passed through Open vSwitch caused a size calculation to wrap, so packet data landed outside its buffer.
- PPPoEject — in PPP over Ethernet handling. Code retained a pointer into a network buffer while calling a device routine that could free and move that buffer, producing a use-after-free when later writes accessed freed memory.
- DiagSpill — in SCTP reporting code. A 16-bit counter tracking connection endpoints wrapped at 65,536, causing the reporting code to copy a full list into no space and write roughly 8 MiB past the buffer.
The cyber insurance questionnaire just landed. Now what?
SOC 2, HIPAA, insurance renewals - someone has to own security strategy. Nubivance provides fractional CISO leadership without the full-time salary.
Get a security leadTechnical reach: local, narrow remote, and container considerations
All four are memory-safety bugs in various parts of the kernel's networking stack. Manizada reported that the underlying mistakes are old — ranging from 10 to 21 years — and that two of the flaws, DirtyAH6 and DiagSpill, can be triggered over the network but only in narrow scenarios and primarily to crash the system.
DirtyAH6 can crash a host that functions as an IPv6 router or gateway and that adds an IPsec Authentication Header in transport mode. Manizada reached remote root with DirtyAH6 only in his own lab, and only after shaping the target's memory first; he wrote that doing that from a remote position alone "looks extremely difficult," though he did not rule it out. DiagSpill can crash a host only when certain SCTP options — which are off by default — are enabled; Manizada said he sees no path to remote root for DiagSpill, even with perfect memory shaping.
Manizada also noted that the flaws could, in theory, allow escape from a container, but he did not build an exploit to demonstrate that.
Patch status and short-term mitigations
The mainline Linux kernel versions that first include the full set of fixes are:
- 5.10.270
- 5.15.221
- 6.1.188
- 6.6.157
- 6.12.109
- 6.18.50
- 7.2.4
Most users run distribution-supplied kernels (for example Debian, Ubuntu, Red Hat, or SUSE) which use their own versioning and backporting schedules; administrators should check their distribution's security advisory to confirm the update includes all four fixes rather than simply matching the upstream version numbers listed above.
If an immediate kernel update is not possible, two mitigations reduce exposure:
- Turn off unprivileged user namespaces. That closes the ordinary-user path to DirtyAH6, TUNderflow, and PPPoEject. It does not mitigate DiagSpill, nor does it stop a container or process that already has network-admin privileges.
- Disable the affected features if they are not needed: IPv6 AH6, TUN/TAP devices, PPPoE, and SCTP.
Manizada recommends patching rather than relying on feature disablement, because other paths to the same flaws may exist.
How this sits with ongoing AI-assisted discovery
Manizada said he found the four flaws with an AI-assisted process that builds a map of how the kernel handles memory and reasons about layout. The kernel's fix for DirtyAH6 carries an "Assisted-by" line that credits his custom AI tooling. This disclosure follows a run of Linux kernel privilege-escalation flaws disclosed through 2026, several found with help from large language models; Manizada disclosed a related Open vSwitch flaw, OVSwrap, in July. In his write-up, he said this batch likely ends the public phase of his AI-assisted bug hunting.
For now, the concrete action is straightforward: confirm that your kernel — or your distribution's kernel package — carries the fixes listed above, or apply the temporary mitigations until you can patch. The presence of public exploit code raises windows of opportunity on unpatched, shared systems; closing those windows remains the immediate priority.
