Linus Torvalds has told Linux kernel contributors that AI-assisted code review is here to stay — and that resisters can take their objections elsewhere.
The flashpoint was Sashiko, an opt-in, multi-stage AI tool that analyzes kernel patches before they're merged. Its creators claim it catches 53.6% of bugs in patches that already passed human review, with a false positive rate estimated at under 20%. Developer Laurent Pinchart proposed filtering Sashiko's output through a human triage step, citing Software Freedom Conservancy guidelines on AI-generated code. Roman Gushchin, one of Sashiko's authors at Google, pushed back, arguing that triage would gut the tool's value. Torvalds agreed, and then some: he posted to the kernel mailing list that Linux "is not one of those anti-AI projects" and that anyone with a problem can fork the project or walk away.
The statement matters because Torvalds carries unusual authority in open-source — his word on tooling shapes what thousands of contributors treat as acceptable practice. His 2024 position was that AI was overhyped; his current one is that it's "clearly a useful tool" and no longer in question. That reversal, made publicly and bluntly, gives AI-assisted review a kind of institutional legitimacy in the Linux ecosystem that a quieter endorsement would not.
Other major open-source projects — Gentoo Linux, Curl, and Ghostty among them — have moved in the opposite direction, restricting or banning LLM-generated contributions over concerns about low-quality submissions. Torvalds isn't addressing code generation, which is the source of most of that friction; Sashiko only comments on patches and takes no autonomous action. That distinction is doing a lot of work in his argument, and critics of a broader AI-in-open-source policy will notice it.