Linus Torvalds rarely leaves room for interpretation, and he didn’t here. Weighing in on the kernel’s code review process, he declared that “Linux is not one of those anti-AI projects,” and that anyone unhappy with that can “do the open-source thing and fork it. Or just walk away.” His framing is deliberately deflationary: AI is “a tool, just like other tools we use,” and “clearly a useful one.”
Not an endorsement of hype — a refusal to ban
It’s worth being precise about what Torvalds actually said, because the headlines will flatten it. He isn’t promising AI writes the kernel now, nor is he vouching for the quality of any particular model’s output. He even hedges that its usefulness “may not have been that clear” earlier on. What he’s doing is narrower and, in some ways, more consequential: he’s refusing to let AI-assisted contributions be rejected on principle. The line in the sand isn’t “use AI” — it’s “you don’t get to veto a patch just because a tool touched it.”
That distinction matters because the kernel’s real currency has always been review. Maintainers judge diffs on correctness, style, and maintainability, not on the biography of how they were produced. Torvalds is essentially extending that norm to AI: the patch stands or falls on its merits, and the provenance is not a disqualifier. For a project built on ruthless code review, that’s a coherent position rather than a capitulation.
The culture fight open source has been avoiding
What’s telling is that Torvalds felt he had to say this at all. Anti-AI sentiment in developer communities is real and often principled — licensing and training-data concerns, worries about slop patches, maintainer burnout from reviewing low-effort machine output. By planting his flag as “top-level maintainer,” Torvalds is signaling that these debates won’t be settled by quietly gatekeeping submissions. He’d rather have the argument in the open, with the fork button as the pressure valve.
The “fork it or walk away” framing is pure Torvalds theater, but it also encodes how open source actually resolves disputes: through exit, not through moderation. He isn’t ordering anyone to like AI. He’s saying the project’s direction is set, and dissenters have the same recourse they’ve always had.
Why this echoes beyond the kernel
When the maintainer of the most-deployed codebase on Earth treats AI tooling as simply allowed, it sets a reference point every other project will be measured against. Expect more maintainers to be asked, pointedly, where they stand. The harder, unglamorous questions remain unanswered — how to keep AI-generated slop out of review queues, how to attribute and license machine-assisted work — and Torvalds’ pragmatism doesn’t solve them. But he’s made one thing clear: in the kernel, the debate is over whether AI is welcome. It’s now about whether the code is good.