Linus Torvalds Sets Linux Kernel's Official AI Policy

Key takeaways

1. The official position, settled by the supreme authority. Worth keeping verbatim: "Linux is not an anti-AI project", and Torvalds "puts his foot down" as top-level maintainer. Disagreement isn't censored, but it is outside the project's policy: "do the open source thing, fork it — or leave". A rare case where a major open source project's AI doctrine is explicitly stated.
2. The thesis in one sentence: "AI is a tool, and clearly a useful tool.". Usefulness is no longer an open question ("it may not have been a year ago, it isn't now"). Torvalds separates two debates: (a) questions still open — notably "what the AI economy will look like"; (b) the settled question — "is it useful?". Jab: "anybody who doubts that clearly hasn't actually tried it".
3. Honesty about the cost — AI is "painful". Notable concession: maintainer overload, and AI "keeps finding embarrassing bugs". But the response isn't the ostrich approach ("La La La, I can't hear you"): it's making sure LLMs _help_ maintainers instead of harming them (the whole point of the Patchwork↔Sashiko thread).
4. Governance: non-coercion + intransigence. Key formula: "nobody is forced to use it, but I will very loudly ignore those who try to prevent others from using it". Freedom to use it is protected; an anti-AI veto imposed on others is not.
5. "Natural intelligence isn't always all that great either.". A rebuttal to the charge that AI is imperfect: whoever points at its flaws "had better also point at the mirror". Worth keeping to defuse the human/AI double standard.
6. Ideological boundary: technology, not activism. The kernel "has always been and will remain about technology"; the social angle of open source is a side benefit, not the point; "this is NOT a social warrior project". We do open source "because it results in better technology, not for religious reasons". Program-punchline: "decisions based on technical merit first — not fear of new tools".
7. Context: Patchwork ↔ Sashiko. The concrete debate is about linking Patchwork (the kernel's email-based patch tracker) with Sashiko (an LLM maintainer-assistance tool). The adverse objection: doing so would make "Sashiko's purpose — helping maintainers — unreachable" if every use case had to be complicated to avoid LLMs. Torvalds reframes it: the real issue isn't how to complicate things, it's accepting that LLMs help.
8. Watch use. a doctrinal position statement (not reporting), with high symbolic value — two pillars of open source (Torvalds here, ESR in [[raymond-llm-coding-empowering-2026-07-08]]) display in July 2026 a pro-LLM pragmatism against the "anti-AI" camp. Useful for the thesis that "coding AI adoption plays out in community governance, not just tooling", and as an authoritative counterpoint to disillusionment narratives.