# Linus Torvalds Tells AI Critics to Fork Off: Linux Is Not Anti-AI > Source: > Published: 2026-07-18 10:30:00+00:00 Linus Torvalds sent a message to the Linux kernel mailing list on July 15 that left very little room for interpretation: Linux is “not one of those anti-AI projects,” and contributors who disagree are free to fork the code or walk away. For a community that has been quietly fracturing over AI tools in kernel development for the better part of a year, it was a clean, loud declaration. The debate is over. ## What Actually Changed This lands harder when you remember where Torvalds stood twenty months ago. In October 2024, he dismissed “basically 90 per cent of AI” as “marketing hype” — a quote that got cited widely, because it reflected a real skepticism held by many experienced engineers. The technology had not yet earned credibility in their world. It has since. And the thing that appears to have shifted the argument is not the general AI conversation but one specific tool: [Sashiko](https://www.phoronix.com/news/Sashiko-Linux-AI-Code-Review), an AI code-review system built by Google engineers and open-sourced in March 2026. Sashiko tested against 1,000 recent upstream kernel patches — all already merged after passing human code review — and found bugs in roughly 53% of them. Not toy examples. Real bugs in the main Linux tree, caught after the fact. When AI is finding more than half the bugs that experienced kernel engineers missed, [“90% hype”](https://www.tomshardware.com/tech-industry/artificial-intelligence/linus-torvalds-reckons-ai-is-90-percent-marketing-and-10-percent-reality) becomes a hard stance to hold. ## The Friction Behind the Policy Torvalds’ statement was not a calm declaration from a settled debate. The community had been generating real conflict. The most pointed example: kernel developer Krzysztof Kozlowski objected when Sashiko attached a “Reviewed-by:” tag to a patch. That trailer is not a rubber stamp in Linux workflow — it signals that a specific person has reviewed the change and accepts some accountability for it. A bot cannot make that statement. Kozlowski’s objection was procedurally sound. There was also volume. Linux 7.1-rc5, released in May 2026, was unusually large — Torvalds himself noted it was “quite a bit bigger than rc5’s have traditionally been.” Part of that pressure came from AI-generated bug reports flooding maintainers, duplicating work and eroding release discipline. Media subsystem maintainer Laurent Pinchart argued that contributors should not be forced to respond to automated reviews, particularly where signal-to-noise ratios are poor. Torvalds acknowledged all of this. AI is, in his words, “a somewhat painful tool” — it increases maintainer workloads and tends to surface embarrassing bugs. His response was not to block the tool but to demand better tooling: ensure LLM systems help maintainers rather than overwhelm them. ## What the Policy Actually Says Read it carefully, because precision matters here. Torvalds is not saying AI-generated patches get a free pass. Technical merit remains the standard it has always been. A bad patch is a bad patch, regardless of how it was written. What the policy protects is the right to use AI tools at all. Contributors can use Cursor, Claude Code, GitHub Copilot, or any other AI assistant when writing patches. Reviewers cannot reject a contribution on the grounds that AI was involved. Maintainers cannot enforce an ideological objection. “I will absolutely put my foot down as the top-level maintainer,” Torvalds said. The practical result is already visible: subsystem maintainers are adapting independently. Theodore Ts’o, who maintains ext4, found Sashiko’s false-positive rate acceptable and expanded its coverage. Media subsystem maintainer Laurent Pinchart has not. That local variation is probably the most workable outcome for a project as broad as the Linux kernel. ## Why This Matters Outside Linux The kernel is not just one open-source project. It is, by influence, the flagship. When [Torvalds draws a line](https://www.theregister.com/ai-and-ml/2026/07/15/linus-torvalds-tells-ai-haters-to-fork-off/) on AI policy, projects across the ecosystem take notice. Compare this to [ESLint’s approach](https://eslint.org/docs/latest/contribute/ai-policy) — which welcomes AI contributions but requires disclosure and does not compensate AI-primary work — and the kernel is now noticeably more permissive. That Overton window shift matters. Projects that were holding out on permissive AI policies now have cover. Projects that were already permissive feel validated. And developers who have been quietly using AI tools while nervously wondering if they were violating community norms have their answer. The kernel community’s debate about AI was never going to resolve itself incrementally. It needed someone to draw the line. Torvalds drew it.