Security firm **DepthFirst** has unveiled a significant use-after-free vulnerability within the **Linux kernel**'s AF_UNIX socket subsystem. This flaw, identified as **CVE-2026-80521** (CVSS score: 7.8), presents a critical risk, allowing for container escape and subsequent root privilege escalation on the host. ### Unpatched Ubuntu Releases at Risk The vulnerability was patched upstream on August 6 in **Linux kernel** versions 7.2 and 7.1.10. However, **Ubuntu** has not yet integrated this fix into its 26.04, 24.04, or 22.04 LTS releases, leaving many systems exposed. **DepthFirst** has publicly released exploit code specifically targeting **Ubuntu 26.04**, demonstrating the practical threat. **Ubuntu**'s security tracker lists the **Linux** package on 26.04 as "vulnerable, work in progress." The 24.04 and 22.04 releases are also affected, including kernel packages deployed for **AWS**, **Azure**, and **GCP** workloads. As of now, no official fix has been shipped for any of these affected releases. Currently, **CVE-2026-80521** is not listed in **CISA**'s Known Exploited Vulnerabilities catalog, and there are no confirmed reports of in-the-wild attacks leveraging this flaw. ### The Mechanics of the Flaw The vulnerability resides within the kernel's garbage collector for AF_UNIX sockets, which is responsible for cleaning up file descriptors passed between processes via SCM_RIGHTS messages. AF_UNIX sockets facilitate local inter-process communication and are permitted by default in **Docker** and **Kubernetes** seccomp profiles, making the flaw reachable from within a container. A race condition during the garbage collection process allows it to prematurely process new references before the data carrying them has been properly queued. If the collector executes during this narrow window, it can deallocate a portion of a linked socket group without removing its corresponding pointer from a persistent internal list. The subsequent collection pass then attempts to access this freed memory, leading to the use-after-free condition. Crucially, the exploit leverages ordinary system calls that containers are typically allowed to make, effectively bypassing namespace isolation, cgroup limits, and seccomp filtering.  The vulnerable code was initially introduced in kernel 6.10 and subsequently backported to stable branches 6.1 and 6.6. Organizations running an affected kernel are advised to apply the upstream patch directly. While **Ubuntu**'s tracker indicates "work in progress," a definitive date for a distribution update remains unannounced. Neither **DepthFirst** nor **Ubuntu** has published a temporary workaround. **DepthFirst** recommends migrating untrusted workloads to microVM isolation solutions, such as **Firecracker** or **Kata Containers**, which provide each workload with its own kernel, thereby avoiding reliance on the host's shared kernel. ### Discovery and Implications for Container Security **DepthFirst** attributes the discovery of this flaw to its AI model, **dfs-large1**, trained for vulnerability detection, working in conjunction with a human-operated testing harness. The company secured a **Google kernelCTF** slot with the exploit on July 24 and reported the bug to the kernel security team on August 5. The kernel maintainers informed **DepthFirst** that a researcher at **OpenAI** had independently reported the same bug. The **CVE** commit credits kernel-exploitation researcher **Kyle Zeng** as the reporter. This disclosure follows a series of **Linux kernel** flaws in 2026 that have facilitated container escapes. A **futex** vulnerability disclosed in July and a flaw in the kernel's cryptographic subsystem in April also allowed unprivileged users to escalate to root on the host. Both discoveries similarly involved AI-assisted research. **DepthFirst** argues that the advent of AI-accelerated vulnerability discovery has significantly lowered the barrier to container escapes, suggesting that organizations should no longer consider containers a robust security boundary. > "The barrier to escaping containers by attacking the kernel has fallen so significantly that we must assume attackers can do so at will," the company stated. With nearly 5,700 **Linux kernel CVEs** published in 2026—the highest annual total on record, according to **LinuxCVETracker**—the demonstrated exploit and the increasing volume of vulnerabilities underscore the urgency of rethinking container security strategies.