Torvalds cuts it short: the Linux kernel will not be an "AI-free" space

On the LKML, Linus Torvalds cut short the controversy in one sentence - contributors who refuse any code touched by AI only need to fork. What this technically says about how open source will live with LLM.

The context

For a few months now, the question has been raised on the Linux Kernel Mailing List (LKML, the mailing list where kernel patches are discussed): should contributions produced or assisted by generative AI be refused? Some maintainers wanted to explicitly ban patches from LLM (Large Language Models - large models like GPT or Claude). Others, more pragmatic, judged the approach impossible to enforce and above all off-topic: what matters is the quality of the diff, not the way the contributor produced it.

Linus Torvalds, as often happens, eventually waded into the debate.

What Torvalds said

His position, summarized: the Linux kernel is not - and will never be - a "no-AI zone" project. Contributors who absolutely want to work in an environment with no code touched by AI are free to fork the kernel. Or, in the phrase repeated by The Register, to "just walk away."

The gist of his argument:

• A maintainer remains responsible for every line they accept, whether it comes from a human brain, a sed, or an LLM. The review process does not change.
• Banning AI is unverifiable: no one can prove that a patch was written by hand.
• The project's policy is the rule of code, not the sociology of its production.

Why this is interesting for a dev

Two things to remember. First, Torvalds does not comment on the intrinsic quality of code generated by LLM - he just reminds us that the filter is the review. A poor patch is rejected whether it comes from an intern or from GPT. Second, he buries the idea of an AI-free certification in a free project of this size: Linux runs on several billion machines, its contributors are everywhere, enforcing it would be fiction.

It's also a political stance. On many FOSS (Free and Open Source Software) projects, the question becomes inflammatory - see below the case of the "Haskell defector" that the Haskell world pilloried for defending the use of AI. Torvalds refuses to enter this framework.

Key takeaways

• Linux will not have an anti-AI policy; maintainers remain the sole judges of their patches.
• The rule remains: the review is the only filter that counts.
• Those who want an "AI-free" kernel have the constitutional right of free software: the fork.