Linus Torvalds has pushed back against attempts to keep AI-generated review comments out of the Linux kernel development process, saying the project will not treat large language model tooling as disqualifying on its own.
The argument surfaced on the Linux kernel mailing list during a discussion about Sashiko, an opt-in code review system for mailing lists that analyzes proposed kernel patches and posts comments. According to the Sashiko project page, the tool uses a multi-stage review process and has found 53.6% of bugs in submitted patches. The page says that figure compares favorably with human review because the patches had already been reviewed by people. It also says the false-positive rate is harder to measure, but estimates it within 20%.
Sashiko does not merge code, reject patches, or change the tree by itself. It comments. That distinction matters in a kernel process built around mailing-list review, maintainer judgment, and a long institutional memory for bad patches arriving with confident explanations.
Developer Laurent Pinchart argued that Sashiko output should be checked before comments reach patch authors. He pointed to the Software Freedom Conservancy’s recommendations on LLM-backed generative AI, which set out cautions for projects dealing with AI-generated work.
Roman Gushchin, a Google engineer and one of Sashiko’s creators, responded that pre-screening would weaken the tool’s usefulness. He also characterized Pinchart’s position as anti-LLM. Torvalds agreed with that assessment and made clear that, at least for Linux, opposition to AI tools as a category will not set project policy.
“I realize that some people really dislike AI, but this is an area where I'm willing to absolutely put my foot down,” Torvalds wrote on the mailing list. “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.”
Torvalds framed AI review as another engineering aid, rather than a political test for participation. “AI is a tool, just like other tools we use. And it's clearly a useful one,” he wrote, adding that whatever uncertainty existed about usefulness a year ago is no longer the point for him.
The stance is sharper than Torvalds’ earlier public skepticism about AI marketing. In 2024, he dismissed much of the surrounding pitchwork as hype, according to The Register. His latest comments do not endorse every LLM-generated contribution or every vendor claim. They do say that a tool which finds real patch problems should not be blocked because it uses AI.
Torvalds also said Linux is about improving technology rather than acting as what he called a “social warrior” project. He wrote that Sashiko “keeps finding embarrassing bugs” and said he would ignore arguments aimed at preventing others from using it.
The practical policy signal is blunt: Linux maintainers may argue about the quality of a specific review comment, the noise rate of a tool, or whether a warning is correct. Torvalds is not accepting a blanket anti-AI rule for the kernel review pipeline. His jab at human reviewers landed in the same thread: “it's not like natural intelligence is always all that great either.”
This story draws on original reporting from Tom's Hardware.
