Linus Torvalds backs AI coding in Linux: fork it, or walk away

Lead

AI coding tools are no longer just a productivity debate. In major open source projects, they are becoming a governance question: can one group of contributors prevent others from using them? In the Linux kernel community, Linus Torvalds has answered bluntly. Linux, he argued on the kernel mailing list, is not an anti-AI project; those who cannot accept that can follow the open source route and fork it, or simply walk away.

The immediate trigger was a discussion around Sashiko, described by its creators as an agentic Linux kernel code review system. According to the material, Sashiko can identify a portion of bugs that human developers later fix, but it can also produce false positives, creating extra work for maintainers who must decide whether a report is meaningful or noise.

Key points

• Torvalds rejects a blanket AI ban. His position is not that everyone must use LLM-based tools, but that opponents should not stop others from using them in Linux development.
• The issue goes beyond generated code. Tools like Sashiko are not merely writing patches; they are attempting to assist review, bug discovery and maintenance workflows.
• False positives matter. Even a useful system can become disruptive if it sends maintainers too many incorrect reports. In a project as complex as the Linux kernel, attention is a scarce resource.
• The standard is technical merit. Torvalds frames AI as another engineering tool. It should be evaluated by its practical usefulness, not rejected because it is new or controversial.
• Open source norms are under pressure. Some groups argue that contributors should be able to reject LLM-generated AI systems. Others in the broader community have gone further, using hostile prompt-injection tactics to deter vibe coding bots.

Why it matters

Torvalds' stance carries weight because Linux is one of the most important and demanding open source projects in the world. If the kernel community accepts that AI tools can be part of development, the question for other projects may shift from whether AI is allowed at all to how it should be governed.

That does not mean AI gets a free pass. The harder work is defining operational rules. Should AI-assisted submissions be labeled? Should automated bug reports be rate-limited? What false-positive level is acceptable? Can maintainers refuse reports from specific tools or bots? These are practical policy questions, and they are more useful than a simple pro-AI or anti-AI slogan.

Torvalds' argument is ultimately an engineering one: tools should not be excluded because of their identity, but they should not be trusted just because they are fashionable either. AI must prove its value through better code, faster review, stronger testing or more effective bug discovery. At the same time, anti-AI absolutism has limited leverage in a project whose culture prizes technical results and the freedom to fork.

For Linux, AI is unlikely to replace maintainers' judgment. But it is likely to appear more often in patches, review workflows, testing and bug reports. The real challenge is making these systems increase signal rather than amplify noise.

Source: Ars Technica AI