AI Assists Researcher in Developing Linux Kernel Root Exploit for CVE-2026-53264
A researcher leveraged AI to discover and exploit a critical use-after-free race condition in the Linux kernel's traffic-control subsystem, enabling local privilege escalation to root.
A significant vulnerability within the Linux kernel's traffic-control subsystem, identified as CVE-2026-53264, has been exploited by a researcher to achieve root-level privileges on CentOS Stream 9. The flaw, a use-after-free race condition, allows an unprivileged local user to escalate their access on a compromised system. This discovery and the subsequent development of a functional exploit were notably accelerated by the use of artificial intelligence (AI) tools, according to researcher Lee Jia Jie.
The technical details of the exploit reveal that it requires specific kernel configurations, including enabled user namespaces and the CONFIG_NET_ACT_GACT and CONFIG_NET_CLS_FLOWER options. Furthermore, the exploit relies on a kernel-specific return-oriented programming (ROP) chain with hardcoded offsets, which narrows its immediate applicability but does not diminish the urgency for patching given the public availability of the exploit code. The vulnerability lies in the kernel's handling of traffic-control actions, where concurrent operations can lead to a race condition, leaving a thread attempting to read an object that has already been freed.
The upstream fix for CVE-2026-53264 was integrated into the Linux kernel on June 1, 2026, and has since been backported to various stable kernel branches. Vulnerable kernel versions begin with Linux 4.14, with fixed releases including 5.10.259, 5.15.210, 6.1.176, 6.6.143, 6.12.94, 6.18.36, and 7.0.13. The mainline fix was merged into the 7.1-rc7 branch. Users are advised to ensure their distribution's kernel packages incorporate these fixes, rather than solely relying on upstream version numbers.
As of July 28, 2026, the vulnerability has not been listed in CISA's Known Exploited Vulnerabilities (KEV) catalog, nor have there been official reports of widespread exploitation in the wild. However, the public release of exploit code by STAR Labs, targeting CentOS, significantly raises the stakes for systems that remain unpatched and meet the exploit's prerequisites. The exploit's success rate was reported as 100% in the researcher's tests, with execution times ranging from nine to 111 seconds on a laptop, though these figures have not been independently verified.
Lee Jia Jie detailed how AI assisted in multiple stages of the vulnerability lifecycle, from initial discovery and the creation of a Kernel Address Sanitizer (KASAN) proof-of-concept to the optimization of the race window. Despite the AI's contribution, Lee emphasized that human judgment remained critical throughout the process, noting that AI still possesses significant limitations in reasoning and can have blind spots. The precise nature of the AI system and its interaction with the researcher were not fully disclosed, making it difficult to quantify the AI's exact contribution versus human direction.
The exploit's methodology involves creating isolated user and network namespaces to gain CAP_NET_ADMIN privileges locally. It then triggers the vulnerable path using a specific traffic control queueing discipline (qdisc) and filter. Operations like timerfd and epoll are used to widen the race window, allowing for the reclamation of the freed object. The ROP chain ultimately overwrites the core_pattern setting, a critical kernel parameter that dictates how core dumps are handled, enabling the execution of arbitrary code.
To achieve privilege escalation, the exploit places a copy of itself into a memory file descriptor (memfd) and intentionally causes a child process to crash. This triggers the Linux kernel to execute the memfd-backed binary as the root core-dump handler within the initial namespace, thereby granting root privileges. The exploit's reliance on fixed gadget offsets means it may require modification for different kernel packages or newer builds, potentially limiting its universal applicability but not its threat to vulnerable, unpatched systems.
The disclosure process also saw Kyle Zeng credited in the upstream patch for reporting the vulnerability shortly before the TyphoonPwn 2026 competition. Lee Jia Jie independently discovered the same flaw and subsequently published his detailed technical analysis and exploit code. Distribution status for the fix varies, with Debian and Ubuntu showing progress in patching, while SUSE lists the issue as pending across multiple products. The CVSS score assigned by the Linux CNA is 7.8, though SUSE's assessment is lower, focusing solely on availability impact.
This new report provides further technical details on the exploitation of CVE-2026-53264, including a diagram illustrating the exploit path from unprivileged namespaces to root execution via core dump overwrite. It also details the specific components used in a successful exploit, such as KASLR bypass and heap reclamation, and mentions additional exploitable bugs found in kernel/events/core.c, including CVE-2026-64300.