"Fork it or leave": Linus Torvalds riles up Linux's AI luddites

Linus Torvalds has clashed with Linux contributors over AI, defending Sashiko reviews while dismissing calls to keep AI out.

Linus Torvalds, creator of Linux, has pushed back hard against contributors opposing AI in Linux, suggesting that those who disagree can "do the open-source thing and fork it" or simply walk away.

Apparently, Laurent Pinchart, a long-time contributor to the Linux kernel, thinks that code reviews generated by Sashiko, a generative AI bot that reviews patches submitted to the project, should not be sent directly to the developers who wrote the code.

Maintainers debated whether developers should have the option to opt out of these automated reviews. What if a developer opts out of receiving these automated messages, but the AI finds a critical bug that the maintainer subsequently misses? To justify his stance, Pinchart pointed to the Software Freedom Conservancy (SFC)'s June 18 guidelines for LLM-backed tools, which advise open-source projects to respect developers who reject AI.

Another contributor, Roman Gushchin, thought it's pointless to have a human developer manually verify and forward every single AI review because, after all, AI is supposed to save the maintainer time, plus the SFC's very anti-AI stance is not something the Linux project shares.

The "AI makes development" faster thing is quite interesting because the media subsystem developers have tried letting the Sashiko post its reviews directly to the main developer mailing list. The experiment failed when the AI produced the usual word salad of hallucinated issues, which confused developers, who then escalated these AI comments to human maintainers, creating more work for the maintainers.

Linus agrees that Linux is not "anti-AI" and has no patience for developers who put their heads in the sand and sing "La La La, I can't hear you" at the top of their lungs. He declared that he will absolutely put his foot down as the top-level maintainer to keep AI tools in the development process because they clearly help people find bugs. Torvalds pointed out that those who criticize AI shortcomings should look in the mirror because natural human intelligence frequently makes embarrassing mistakes of its own and that Linux does not operate as some sort of "social warrior project".

Linus previously called AI "a bubble and a revolution" in late 2025, pointing out that while the tools work incredibly well for kickstarting new projects, they fail completely at long-term code maintenance. In May, he also expressed immense anger when people flooded the kernel with identical, low-quality bug reports from LLMs. To stop this, he stripped those reports of their confidentiality, forcing researchers to stop dumping a giant low-quality wall of text into private security channels and start doing real verification work.

The open-source world faces a huge problem at the moment. AI models have gotten ridiculously good, which is great because, given the right context, they help write mostly-correct code faster, but it's also "bad" because they lower the barrier so much that just about anybody can generate, for example, a bug report or a patch without knowing whether it's correct or even understanding it.

Earlier this month, the popular open-source 2D/3D game engine Godot, which powers indie hits like Brotato, updated its contribution policy to ban pull requests from AI entirely. The RPCS3 PlayStation 3 emulator team did something similar in May 2026, telling developers to stop submitting AI code that they do not understand.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Thursday 16 July 2026 at 12:38 pm AEST (my time).

News posts: 2023 5,800+ | 2024 5,700+ | 2025 5,700+ | 2026 (to end of June) 2,475

RIP Matrix