Torvalds Draws a Line in the Sand

Linus Torvalds has had enough. In a blistering post on the Linux kernel mailing list, the project's creator and top maintainer told anti-AI absolutists exactly where they can go: "Fork it. Or just walk away."

The statement came in response to a thread discussing Sashiko, an "agentic Linux kernel code review system" that uses AI to autonomously find bugs. According to its creators, Sashiko can independently detect 53.6% of bugs that would later be fixed by human developers. The trade-off? A false positive rate "well within [the] 20% range," meaning maintainers may receive bogus bug reports.

Some contributors pushed back, citing the Software Freedom Conservancy's recent declaration that the open source community "should support, not just tolerate, those who outright reject LLM-gen-AI systems." Torvalds was unmoved.

"We're not forcing anybody to use [LLM tools], but I will very loudly ignore people who try to argue against other people from using it," Torvalds wrote. He added that he is "willing to absolutely put my foot down" in support of AI-assisted development.

The Sashiko Debate

Sashiko isn't just another code linter. It's an agentic system that scans kernel commits and independently submits bug reports. The 53.6% detection rate is significant — nearly half of all bugs that eventually get fixed are caught by the AI before a human even sees them. But the 20% false positive rate means one in five alerts is noise.

In a project as large as the Linux kernel — with thousands of active maintainers and millions of lines of code — the burden of triaging false positives is real. One mailing list poster argued that maintainers shouldn't be forced to wade through AI-generated spam. Torvalds disagreed, framing the issue as one of efficiency versus purity.

"If you don't want to use AI tools, don't. But don't tell others they can't," he said.

Torvalds' History with Controversy

This isn't the first time Torvalds has shut down ideological debates with blunt force. In 2018, he took a break from kernel maintenance to work on his "people skills," but his core philosophy remains: technical merit trumps all. AI tools that demonstrably improve bug detection are welcome, regardless of how they were built.

The Linux kernel has long accepted patches generated by non-human tools — compilers, static analyzers, and automated refactoring scripts. LLM-generated code is simply the next iteration. Torvalds sees no principled distinction between a patch written by a human and one generated by an LLM, as long as it passes review.

Practical Implications for Kernel Contributors

If you contribute to the Linux kernel, here's what this means:

  1. AI-assisted patches are welcome. You can use tools like Sashiko, GitHub Copilot, or any LLM to help write or review code. Just make sure your patches meet the usual standards.

  2. False positives are part of the deal. If you're a maintainer, expect more automated bug reports. Torvalds has signaled that he'll side with AI tooling, so adapt or filter aggressively.

  3. No opt-out for anti-AI contributors. If you refuse to work with LLM-generated code, you're free to fork the kernel. But within mainline Linux, there is no carve-out.

The Bigger Picture

Torvalds' stance aligns with a growing trend in large open-source projects. The Python Software Foundation recently debated similar issues, and the Apache Software Foundation has issued guidelines for AI contributions. But Torvalds' directness is unusual — most project leaders tiptoe around the controversy.

By threatening a fork, Torvalds is calling the anti-AI crowd's bluff. Forks require sustained effort, and few have the stamina to maintain a kernel-compatible alternative. In practice, the decision is likely to accelerate AI adoption in kernel development.

What Developers Should Do

If you're a kernel contributor, update your workflow now. Try Sashiko for automated review, or integrate an LLM into your patch submission pipeline. If you're a maintainer, set up email filters for Sashiko's reports — the signal-to-noise ratio is still better than no signal at all.

For those who can't stomach AI in open source, Torvalds has made the path clear. The source is open. The code is yours. Fork it.