GhostLock: Public Exploit Grants Root in 5 Seconds on Linux…

CVE-2026-43499 has existed in the Linux kernel for 15 years. The public proof-of-concept requires only a local user to obtain root in roughly five seconds with 97% reliability.

On July 7, 2026, Nebula Security's VEGA research team published a working exploit for CVE-2026-43499, a use-after-free vulnerability in the Linux kernel's rtmutex subsystem present since May 2011. The proof-of-concept allows any local user to gain root privileges in approximately five seconds with 97% reliability, and enables escape from Docker and Kubernetes containers without requiring special capabilities. Combined with a Firefox browser vulnerability, it has already demonstrated a complete remote-to-root chain on Android.

Key Takeaways

The bug resides in the remove_waiter() function of the priority-inheritance rtmutex subsystem: it uses the current pointer instead of waiter->task, leaving a dangling reference to already-freed kernel stack memory.
The public proof-of-concept achieves root in roughly five seconds with a 97% success rate; the remaining 3% results in kernel panic, according to tests on the kernelCTF environment reported by TechTimes.
The exploit escapes containers without privileged capabilities, invalidating the isolation model underpinning cloud-native environments, CI/CD runners, and shared hosting.
Google's kernelCTF program awarded a $92,337 bounty for the discovery, confirming the technically proven severity of the vulnerability.

The Mechanism: A Use-After-Free Hidden for 15 Years in the FUTEX Path

The vulnerability triggers through the FUTEX_CMP_REQUEUE_PI proxy path of the futex subsystem. During waiter cleanup in the remove_waiter() function, the code operates on the current pointer instead of waiter->task. The result is a dangling pi_blocked_on reference pointing to a kernel stack region that is subsequently freed.

The exploit reclaims the freed stack frame via the PR_SET_MM_MAP system call, plants a forged waiter structure, and triggers a controlled write on the inet6_protos function pointer. From there, the researcher uses ROP gadgets in the CPU entry area and the DirtyMode technique to modify core_pattern permissions, completing the privilege escalation entirely from userspace. The CONFIG_FUTEX_PI configuration, enabled by default in every major distribution kernel, is the sole prerequisite.

"The only hard requirement is CONFIG_FUTEX_PI, which is enabled by default in every major distribution kernel. There is no privilege, capability, or namespace requirement. An unprivileged shell — the kind any web app process runs as — is enough to trigger it." — Falcon Internet

Why the CVSS 7.8 Score Is Misleading for Modern Architectures

The CVE-2026-43499 record reports a CVSS 3.1 score of 7.8 out of 10, classified as HIGH but not CRITICAL. The vector AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H explicitly indicates the attacker must already possess local access to the machine. TechTimes notes this score "reflects a specific thing: the attacker must already have local access."

However, the classification ignores the reality of contemporary attack chains. Nebula Security demonstrated IonStack: a chain linking the Firefox IonMonkey sandbox escape vulnerability, CVE-2026-10702, to GhostLock to achieve full control of Android devices with a single tap on a malicious link. In this configuration, the local-access prerequisite is eliminated entirely. Firefox patched CVE-2026-10702 in version 151.0.3 on June 2, 2026, but the demonstration remains valid as a proof of concept for the entire risk class.

The CVSS scoring problem is not technical but semantic: the system measures a single attack step, not the real exposure surface. For multi-tenant environments, CI/CD platforms, GitHub Actions runners, and Linux workstations, the boundary between "local" and "remote" dissolved years ago. A process compromised through any sandboxed application immediately becomes a local vector.

Container Isolation Collapse and Vulnerable Versions

The exploit works from inside a container with standard access to the futex subsystem, without requiring CAP_SYS_ADMIN, CAP_SYS_PTRACE, or analogous capabilities. According to TechTimes, the escape "succeeds from inside a container without requiring any special capability or elevated privileges." Falcon Internet adds that operation is guaranteed "from inside a container running a standard kernel with full futex access."

This property is not an exception but a structural consequence of Linux kernel design, where namespaces and cgroups delegate fundamental security to the kernel itself. When the kernel yields, all isolation mechanisms built atop it become decorative.

The National Vulnerability Database lists affected versions from 2.6.39 through 7.0.4, with partial patches in the 6.1.175, 6.6.140, 6.12.86, 6.18.27, and 7.0.4 series. The upstream fix commit is 3bfdc63936dd, which modifies remove_waiter() to correctly use waiter->task. A subsequent fix, CVE-2026-53166, addresses a null-pointer dereference edge case introduced by the initial patch.

What to Do Now

The patching situation varies significantly across distributions. AlmaLinux has released patched kernels: kernel-4.18.0-553.141.2.el8_10, kernel-5.14.0-687.23.2.el9_8, and kernel-6.12.0-211.31.2.el10_2. Red Hat Enterprise Linux confirmed versions 6 through 10 as affected, with patches in preparation per bulletin RHSB-2026-010. Ubuntu 24.04 LTS, 22.04 LTS, and 20.04 LTS were still vulnerable or undergoing updates as of July 8, 2026; newer releases and cloud-optimized kernels were already patched. KernelCare offers live patches for CloudLinux 7/8/9/10 without requiring a reboot.

Falcon Internet documents a temporary mitigation via a seccomp profile blocking FUTEX_LOCK_PI, FUTEX_WAIT_REQUEUE_PI, and FUTEX_CMP_REQUEUE_PI syscalls. This measure, however, breaks applications that use priority-inheritance mutexes. The compile-time options RANDOMIZE_KSTACK_OFFSET and STATIC_USERMODE_HELPER raise exploitation complexity without fixing the underlying defect.

No infrastructure overlaps link observed activity to active exploitation campaigns in the current state of sources. However, the public availability of the proof-of-concept, combined with 97% reliability and the absence of configuration prerequisites, makes the vulnerability readily weaponizable.

The VEGA Signal: AI in the Discovery Loop

GhostLock was identified using VEGA, an AI-based static analysis tool developed by Nebula Security. The vulnerability survived fifteen years of human kernel code review, including hardening initiatives and dedicated security audits. Its discovery via automated tooling suggests an acceleration in the ability to detect structural defects in legacy codebases of size and complexity comparable to the Linux kernel, currently approximately 30 million lines of code.

The AI-assisted discovery pattern does not imply analogous tools were unavailable previously, but the specificity of the case — a logic bug in task pointer passing within a rare execution path — indicates modern static analysis methods have reached a threshold of practical utility for vulnerability classes previously resistant to automated detection.

Sources

Information verified against cited sources and current as of publication.

Sources

Sources and references