STAR Labs Researcher Uses AI to Turn a Linux Traffic-Control Race Condition Into a Root Exploit
A Singapore researcher disclosed CVE-2026-53264, an AI-assisted Linux kernel use-after-free that escalates local access to root on CentOS Stream 9.
Overview
STAR Labs has published a Linux kernel exploit that turns an ordinary local user into root on the CentOS Stream 9 build it targeted, according to The Hacker News. The flaw, tracked as CVE-2026-53264 with a CVSS score of 7.8, is a use-after-free race condition in the Linux kernel’s network traffic-control subsystem, and researcher Lee Jia Jie said artificial intelligence helped him find the bug and speed up exploit development.
What We Know
The vulnerability sits in net/sched, the Linux kernel’s packet-scheduling subsystem, and stems from mismatched locking around a shared traffic-control action object, according to The Hacker News: one function reads entries under a read-copy-update (RCU) lock, while another path can free the same entry without waiting for the RCU grace period. That gap creates a window in which the kernel can keep using an object after its memory has already been released. The exploit reaches the vulnerable path through a clsact queuing discipline and a flower filter, and requires unprivileged user namespaces along with the CONFIG_NET_ACT_GACT and CONFIG_NET_CLS_FLOWER kernel options, as The Hacker News reported.
According to The Hacker News, this is local privilege escalation rather than remote code execution, meaning an attacker needs an existing foothold on the machine before any of it applies. In Lee’s own testing, the exploit succeeded in all 10 of his runs, taking between nine and 111 seconds on a laptop running CentOS Stream 9 — figures The Hacker News noted have not been independently reproduced.
Lee Jia Jie works for STAR Labs, a Singapore-based offensive security firm, and found the use-after-free during an internship — his first Linux kernel work, according to Infosecurity Magazine. The exploit was originally prepared for TyphoonPwn 2026’s Linux local-privilege-escalation category, which offered prizes of $70,000, $35,000 and $17,500; Jia Jie drew position eight of 11, but the category closed after three winners were selected, so his entry was never demonstrated at the event, Infosecurity Magazine reported. The upstream patch credits Kyle Zeng, who uses the handle KyleBot, as the original reporter; Lee said he found the flaw independently and only later learned that Zeng had reported it shortly before the TyphoonPwn competition, according to The Hacker News.
Lee said AI assisted with vulnerability discovery, producing a Kernel Address Sanitizer proof of concept, and optimizing the timing of the race window, The Hacker News reported. He described the process in a technical write-up, saying the AI-heavy workflow made bug hunting “feel more like I was doing n-day analysis even on new bugs,” according to The Hacker News. He was careful to note the technology’s limits, too: “AI still has many blind spots and lapses in reasoning ability,” he said, adding that human judgment remained necessary throughout the work, per The Hacker News. GBHackers and Infosecurity Magazine both independently reported that the underlying flaw is believed to have existed for approximately two to three years before its disclosure.
The upstream fix landed on June 1, 2026, and has since been backported to several stable kernel branches, according to The Hacker News, which reported that the Linux CNA record lists vulnerable ranges beginning with Linux 4.14 and fixed releases at 5.10.259, 5.15.210, 6.1.176, 6.6.143, 6.12.94, 6.18.36 and 7.0.13, with the mainline fix entering 7.1-rc7. The patch itself defers freeing the affected object until existing RCU readers have finished, closing the race window, The Hacker News reported.
What We Don’t Know
The Hacker News said it found no entry for the flaw in CISA’s Known Exploited Vulnerabilities catalog and no official report of exploitation in the wild as of July 28, 2026. Distribution-level patch status was also uneven at the time of reporting: The Hacker News said Debian listed fixed kernels for its supported stable releases, Ubuntu still marked multiple maintained kernel packages as vulnerable, and SUSE listed the issue as pending across multiple products while separately scoring it 5.5 — lower than the Linux CNA’s 7.8 assessment — using a vector that records only availability impact.
The Hacker News also cautioned that without the underlying model, prompts, service, or full interaction record, the disclosure is difficult to use as a benchmark of AI capability or to separate the system’s contribution from Lee’s own direction and judgment. The outlet said it had asked STAR Labs for further details on the AI system, test environment, and disclosure timeline and would update its reporting with any response.
Analysis
The disclosure lands alongside a broader industry conversation about AI-assisted vulnerability research: it follows earlier work using AI tools such as Google’s OSS-Fuzz to surface hidden flaws in open-source projects, as Infosecurity Magazine noted. Lee’s own framing — that AI can make genuinely new bugs feel like already-known ones to analyze — captures both the promise and the limits security teams are grappling with: faster pattern-matching and proof-of-concept generation, paired with a continued need for deep subsystem expertise to separate real attack paths from AI-generated noise.