Linus Torvalds has taken a blunt stance in the growing debate over AI-assisted programming: the Linux kernel will not be an anti-AI project. In a post to the kernel mailing list, Torvalds pushed back against calls for blanket bans on large language model–generated contributions, saying maintainers should be free to use AI tools and that those opposed can “fork it. Or just walk away.” His comments come amid discussions about automated code-review agents and their role in flagging bugs and generating suggested fixes for one of the world’s largest open source projects.
Torvalds frames the issue as technical, not ideological
Torvalds positioned his support for AI-assisted workflows as pragmatic and grounded in utility rather than enthusiasm for new technology. He argued that AI should be treated like any other development tool and that criticisms must be weighed against human performance. Torvalds noted he will ignore developers who try to block others from using these tools, a position he described with the short, emphatic line: “I will very loudly ignore” such arguments.
That stance reflects Torvalds’s broader approach to kernel stewardship: tolerating a wide range of development styles so long as contributions meet the project’s technical standards. He also shared a personal note about experimenting with “vibe coding” tools on a hobby project earlier this year, describing how he bypassed intermediary steps and leaned on a tool called Google Antigravity to generate an audio visualizer prototype.
Sashiko and the practical limits of automated review
The immediate spark for the thread was Sashiko, an agentic code-review system designed for the Linux kernel. Its creators report that, in tests, the tool independently identified roughly 53.6 percent of bugs that humans later fixed in subsequent commits. However, Sashiko also produces false positives; its maintainers estimate that rate is “well within [the] 20% range.”
That mix—meaningful bug detection paired with a nontrivial false-positive load—fuels the debate over whether maintainers should receive an influx of automated bug reports. Opponents worry such noise could waste reviewer time and obscure truly valuable findings. Proponents argue that even with false positives, automated systems can surface issues humans might miss and speed up triage when used judiciously.
Studies and shifting productivity signals
Objective data on AI-assisted coding remains mixed. A METR study from last year reported that developers using AI tools were 19 percent less productive by some measures, even as those same developers reported feeling 20 percent more productive. The researchers later noted in an early‑2026 update that developer experiences may be changing rapidly and suggested that AI tools seem to be accelerating workflows more now than in early 2025.
Torvalds acknowledged imperfections in AI systems but urged critics to compare AI output to human reviewers, arguing that natural intelligence is not flawless either. His argument hinges on judging tools by technical merit and impact on code quality, rather than endorsing or rejecting AI on principle.
Community friction and safety concerns
Not everyone in the open source ecosystem shares Torvalds’s confidence. Some contributors and organizations have urged respect for developers who prefer to avoid LLM-generated content, emphasizing choice and contributor autonomy. The Software Freedom Conservancy, for example, has advocated supporting those who reject LLM systems and protecting contributors’ self-determination on the matter.
There are also adversarial risks. In May, the maintainer of the jqwik Java testing library added a concealed prompt-injection instruction intended to trick so-called vibe coding bots into discarding jqwik tests and code. That incident highlighted a new class of supply‑chain and tooling attacks that can undermine trust when automated systems ingest and act on external code or prompts without human oversight.
What this means for the Linux kernel and open source
Torvalds’s position is likely to shape how the Linux kernel community treats AI-driven contributions: permissive in principle, with the practical caveat that outputs must meet the project’s quality standards. Allowing maintainers and contributors to use AI tools does not remove the responsibility to vet and review changes. The pushback against absolutist bans suggests a broader trend in open source—accepting new tools while wrestling with their governance, legal, and security implications.
For now, the kernel remains open to AI-assisted workflows. The debate is less about whether such tools exist and more about how to integrate them responsibly: managing false positives, defending against manipulative prompts, and preserving contributors’ rights to opt out. As the tools evolve and empirical evidence accumulates, the Linux community will continue refining norms and practices that balance innovation with codebase integrity.
Torvalds’s message was blunt and clear: the kernel will not be kept pristine by excluding AI, and those who oppose its adoption wholesale are free to leave or fork. The practical test that follows will be whether developers and maintainers can harness AI to improve code quality without overwhelming human reviewers or introducing new vectors of risk.
Source: Ars Technica AI