Linux Kernel Reviewer Steps Back as AI-Generated Patches Flood In
Longtime Linux developer Lorenzo Stoakes is cutting his code-review duties, saying AI-generated patches now make up half or more of what he sees.
Lorenzo Stoakes, a longtime developer of the Linux kernel’s memory-management code, is giving up several of his code-review roles. He says AI-generated patches now make up about half or more of what lands in his inbox. His decision is the latest sign that the world’s most widely used open-source operating system core is struggling with a flood of machine-written contributions.
What happened
On Nov. 10, Stoakes posted a series of four patches to the Linux Kernel Mailing List, the public forum where kernel changes are proposed and debated. Three of them remove his name as a designated reviewer from the kernel’s MAINTAINERS file, the directory that tells contributors whom to send their code to. According to the tech site Xenospectrum, the changes cover core memory management, other memory-management code, memory reclaim, Rust memory management and the Rust allocator. Rust is a programming language that kernel developers have been adopting alongside C.
Stoakes is keeping his maintainer roles, which make him responsible for the areas he oversees directly. He is dropping the broader review work he had done on other developers’ code. In a reply to Andrew Morton, the kernel’s lead memory-management maintainer, Stoakes wrote that he “unfortunately” lacks the bandwidth to offer review in general, according to CommandLinux, which cited reporting by Phoronix. He said he may still review other work from time to time.
He described a steady escalation. Low-quality AI-generated submissions went from appearing occasionally, to weekly, to every few days, and now account for roughly half or more of what he sees. Xenospectrum pointed out that this figure is his own impression, not a measured count.
A more complicated picture
Outlets have described the move somewhat differently. Phoronix’s headline said Stoakes was stepping down from some roles because of an AI-driven patch flood. Xenospectrum noted that Stoakes called the change a “purely administrative measure” and also cited his workload on KVM, the kernel’s built-in virtualization technology. In its view, that means AI cannot be blamed alone.
The fourth patch in the series adds an entry to the kernel’s CREDITS file recognizing his past work as co-maintainer of Transparent Huge Pages (THP), a memory feature that can speed up some workloads. According to CommandLinux, Stoakes had already removed himself from that role in late August, citing his workload. The change was merged on Sept. 2, and developer Zi Yan took over. Xenospectrum said Tom’s Hardware counted THP among the areas he was leaving this week. The new patches, however, only record a credit for that earlier work.
Stoakes has been a critic of low-effort machine-generated code for some time. CommandLinux reported that in January, during review of the kernel’s proposed rules for tool-generated contributions, he asked that maintainers be explicitly allowed to reject such patch series outright.
Why it matters
The Linux kernel runs most of the world’s servers, every Android phone and many embedded devices. It depends on a fairly small group of experienced maintainers and reviewers to check changes before they ship. AI coding tools have made it far cheaper to produce patches and bug reports, but reviewing them still takes skilled human time.
Stoakes is not the only one feeling the strain. Xenospectrum reported these examples from other parts of the kernel:
- Networking: Jakub Kicinski, a networking maintainer, estimated that one-third to one-half of changes merged into the networking development tree for Linux 7.3 were low-priority or cleanup work that appeared to be AI-driven. He called it a rough estimate.
- Wireless: Johannes Berg, the wireless subsystem maintainer, said he would ignore nearly all AI-generated patches from the automated testing system syzbot unless they were obviously correct after a few seconds of checking.
- Staging: The staging tree, which holds immature drivers, reportedly refuses AI-generated patches except for security fixes.
Security reporting has been hit too. Linus Torvalds, who created Linux and still leads its development, has said duplicate AI-generated vulnerability reports made the kernel’s private security mailing list “almost entirely unmanageable,” Tom’s Hardware reported. The project has been moving the handling of such reports to a more public process.
The wider debate
The kernel community is not rejecting AI outright. LWN.net, a long-running publication that covers kernel development, reported in July that Torvalds said “Linux is not one of those anti-AI projects” and told critics they could fork the project or walk away. Some maintainers rely on AI review tools such as Sashiko, an LLM-based system that comments on proposed patches. Others say its output is too noisy or worry that it depends on company-funded access to proprietary models.
Since the 7.0 release, the kernel has asked contributors to label AI-assisted patches with an “Assisted-by” tag. LWN reported that more than 1,200 commits carried the tag by the 7.2 development cycle, while many machine-generated patches did not. Some developers have proposed loosening or removing the requirement. Stoakes is among those who want to keep it, describing the policy as “about empowering maintainers to push back,” according to LWN.
LWN’s reporting also suggested the problem is less about AI itself than about effort. A memory-management series built with an AI model was accepted after its author checked the generated code many times before posting. Another large series drew blunt criticism from Stoakes as far from mergeable. For now, Stoakes’s retreat makes clear that the scarcest resource in kernel development is still the time of human reviewers.
