開源與 AI 開發工具
Linus Torvalds Draws a Line on AI in the Linux Kernel: No Ban, and No Letting Anti-AI Positions Hijack the Project
In a discussion about integrating Sashiko with Patchwork, Linus Torvalds made it clear that the Linux kernel is not an anti-AI project; opponents are free to fork it under open-source rules or leave. This is not blanket approval of AI-generated output, but an application of existing technical evaluation standards to AI tools: the results must be useful, while human maintainers and contributors remain accountable.
Linus Torvalds has taken a hard line on the Linux kernel mailing list over whether AI tools should be excluded from the kernel development process. The discussion arose from an integration between Sashiko and Patchwork. Sashiko is an AI-assisted review system for checking kernel patches, and the controversy centered not on a single patch but on whether Linux should reject LLM tools as a matter of principle.
Torvalds’ position is clear: the Linux kernel will not define itself as an anti-AI project. In his response, he said AI is already a tool like other development tools, and whether it can be useful is no longer the main question. The real issue is how to make LLM tools help maintainers instead of merely adding to their workload. As for those seeking to prohibit others from using AI, he rejected the idea that such a position should dictate the direction of the upstream project.
The phrase “if you don’t like it, fork or leave” is the part most likely to be reduced to a headline, but the technical context matters more than the slogan. Torvalds did not say AI-generated code should be exempt from review, nor did he make vibe coding a kernel development standard. His point was that Linux has always made decisions based on technical merit rather than imposing a default ban simply because a tool carries an AI label. In other words, AI-generated output must still meet the kernel’s review standards and withstand scrutiny over accountability and maintenance costs.
This also reflects the different approaches to AI governance among large open-source projects. Smaller projects may adopt stricter restrictions because of AI-generated bug reports, low-quality PRs, or licensing concerns. The Linux kernel, however, has a more mature maintainer hierarchy and review process, making it more likely to treat AI as a tool that must be brought under control rather than as a contaminant that must be excluded. If Sashiko can consistently identify issues missed by human reviewers, it could become part of maintainers’ workflows. If it merely creates noise, it will likewise be rejected on technical grounds.
The next question is how the Linux kernel will translate this position into procedural details: whether AI review results must disclose their source, which reports may enter public mailing lists, how maintainers should handle duplicate or incorrect reports, and whether contributors must accept full responsibility for AI-assisted output. Torvalds was not declaring that Linux would become an AI-native project. He was drawing a boundary: tools may be used, but they cannot bypass technical review—and an anti-AI position alone is not sufficient justification for blocking them.