Something happened this week that would make any security engineer reach for the strongest coffee available. Between Sunday and Monday, the Linux kernel team published 432 CVEs — common vulnerabilities and exposures — in a single weekend onslaught. That’s more than most organisations see in a quarter. Jan Schaumann, chief information security architect at Akamai Technologies, took to the OSS-SEC mailing list to express what sysadmins everywhere were thinking: “This onslaught really shows it’s not feasible to attempt to prioritise individual kernel changes.”
But here’s the thing that makes this genuinely fascinating — not just alarming. The same week that 432 vulnerabilities hit the wire, Linus Torvalds made his position on AI-generated code in the Linux kernel unmistakably clear: “Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open-source thing and fork it. Or just walk away.”
Two stories, one paradox. The kernel is simultaneously the most vulnerable it’s been and the most open to AI-assisted development. And I find the tension between them more interesting than either story alone.
The CVE tsunami
The 432 CVEs were published across Sunday, July 20 and Monday, July 21. nixCraft flagged the unusual volume on Monday morning, and the reaction from the security community was a mixture of bewilderment and exhaustion.
Schaumann suggested the only realistic response was automated, rolling updates — but even he admitted that’s a luxury most enterprises don’t have. Organisations running RHEL, Oracle Linux, and Amazon Linux with lengthy QA cycles, contractual LTS requirements, and compliance mandates can’t just “update everything on a weekly basis.” They’re stuck trying to triage 432 vulnerabilities by hand.
The Register speculated that AI-assisted bug reports are a likely driver of the volume surge, which wouldn’t be without precedent. Linus himself noted in May that the kernel security mailing list had become flooded with automated reports. The floodgates have opened, and it’s unclear whether that’s a feature or a bug.
The RefluXFS example — AI found this one
One vulnerability from the batch illustrates the paradox perfectly. CVE-2026-64600, dubbed RefluXFS by Qualys Threat Research Unit, is a nine-year-old race condition in the XFS filesystem that allows local attackers to overwrite protected files and gain root privileges. It affects kernels from version 4.11 onwards — meaning virtually every Linux system since February 2017.
Qualys estimates it exposes 16.4 million RHEL and enterprise Linux systems to silent root takeover. Standard defences — SELinux, kernel lockdown, container isolation, KASLR, SMEP, SMAP — don’t block it because the flaw operates at the filesystem allocation layer, below where those protections apply.
Here’s the twist: Saeed Abbasi, head of Qualys TRU, confirmed the discovery emerged from research that integrated Claude Mythos Preview into their manual audit workflow. The AI was tasked with hunting for race conditions resembling Dirty COW. After iterative refinement, it identified the flaw in XFS and generated a functional proof-of-concept. Human researchers then verified every claim and coordinated disclosure.
AI found a vulnerability that had been hiding in plain sight for nine years. That’s the productive side of this story.
The other side — AI writing the code
On the flip side, the volume of AI-assisted patches flooding the kernel has been staggering. According to Memeburn’s analysis, AI-assisted kernel submissions jumped over 2,700% since February 2026, with more than 2,000 AI-generated patches submitted in just the last 45 days.
Torvalds’s response, delivered on the Linux kernel mailing list on July 15, was characteristically blunt. He isn’t backing AI — he’s backing the people who review it. His argument is that writing code is becoming cheap; judging whether it’s safe to merge is where the value has shifted. Code review is no longer just a step in the process. It’s the process.
This puts Linux in direct contrast with other major open-source projects. Gentoo, Curl, and the Ghostty terminal emulator have all moved to limit or outright ban AI-written code over concerns about sloppy, unreviewed patches. Dave Plummer’s 2.5KB Notepad — which I wrote about recently — was itself a protest against the flood of AI-generated pull requests drowning out human contributors.
The Sashiko safety net
Google engineer Roman Gushchin has been working on something called Sashiko — an automatic proofreader for kernel patches. Mailing lists can opt in, and Sashiko scans incoming patches for bugs before any human reviewer looks at them. Gushchin’s team reports internal tests show it catching most bugs a human would find in later review. That number comes from Sashiko’s creators rather than an independent audit, but the concept is sound: if AI is going to write the code, AI should also proofread it before humans give the final verdict.
What this means
I don’t analyse code for a living, but the pattern here is clear to anyone who watches software development. We’re approaching a turning point — where the bottleneck in open-source development has shifted from writing code to reviewing it. The projects that build the strongest review infrastructure will survive the AI flood. Those that don’t will drown in 2,000 low-quality patches a month.
Linux, with its rigorous review process and its community of maintainers who treat code review as a discipline rather than a formality, is positioned to handle this. The question is whether the 432-CVE weekends become the new normal, and whether the people doing the reviewing can scale fast enough.
Schaumann’s despair is understandable. But the alternative — forking the kernel because AI exists, as Torvalds suggested — is not a strategy. It’s a surrender.
**Sources:**
– [The Register: Linux kernel team publishes 432 CVEs in two days](https://www.theregister.com/security/2026/07/22/linux-kernel-team-publishes-432-cves-in-two-days/5276497)
– [BleepingComputer: New RefluXFS Linux flaw lets attackers gain root privileges](https://www.bleepingcomputer.com/news/linux/new-refluxfs-linux-flaw-lets-attackers-gain-root-privileges/)
– [Ars Technica: Linus Torvalds to critics of AI coding in Linux](https://arstechnica.com/ai/2026/07/linus-torvalds-to-critics-of-ai-coding-in-linux-fork-it-or-just-walk-away/)
– [Memeburn: Linus Torvalds Thinks AI Just Made Code Review the Most Important Job](https://memeburn.com/linus-torvalds-thinks-ai-just-made-code-review-the-most-important-job-in-linux/)
