Linus Torvalds, the creator and longtime steward of the Linux kernel, has made his position on artificial intelligence in kernel development unmistakably clear: AI is a tool, it is useful, and anyone who disagrees can fork the project. This blunt directive was posted on the Linux Kernel Mailing List (LKML) in response to discussions about the Software Freedom Conservancy's recent policy statement on AI contributed code. Torvalds' comments have reignited a broader debate across the open-source world about the role of generative AI in software development, but for the kernel community, the matter appears settled.
The kernel community's stance on AI
Torvalds' response came after several developers raised concerns that AI-assisted patches might erode trust or create new kinds of noise in the kernel development process. Some referenced the Zig language project, which has adopted strict policies against AI-generated contributions. Torvalds, however, dismissed such worries as unfounded. In his post, he wrote: 'I realize that some people really dislike AI, but this is an area where I'm willing to absolutely put my foot down as the top-level maintainer.' He added that AI is a tool, just like other tools we use. And it's clearly a useful one. It may not have been that 'clearly' even just a year ago, but it's no longer in question today.
This statement is notable not only for its directness but also because it reflects a consensus that has been growing among the most experienced kernel maintainers. Greg Kroah-Hartman, maintainer of the Linux stable kernel, noted earlier this year that the quality of AI-generated reports and patches has improved dramatically. While months ago the kernel community saw what he called 'AI slop'—obviously wrong or low quality security reports—the situation has now reversed. 'Now we have real reports,' Kroah-Hartman said. 'All open-source projects have real reports that are made with AI, but they're good, and they're real.'
Torvalds echoed that sentiment, acknowledging that AI is far from perfect but arguing that the proper response is not rejection but refinement. 'Yes, it can also be a somewhat painful tool...But the solution is not to put your head in the sand and sing "La La La, I can't hear you" at the top of your voice like some people seem to do.' Instead, he argued, developers should work to make large language models (LLMs) and other AI tools genuinely helpful for maintainers rather than a source of additional burden.
Historical context: Torvalds and decisive leadership
Torvalds' willingness to take a firm, even combative, stance on technical issues is nothing new. Over the decades he has built a reputation for directness, occasionally leading to heated exchanges on the mailing list. Yet his approach has also been widely credited with keeping the kernel development process focused and efficient. His decision to embrace AI aligns with a pragmatic philosophy: if a tool demonstrably improves kernel quality or developer productivity, it should be used. This is consistent with his past championing of new technologies such as Git, which he created after becoming dissatisfied with existing version control systems, and more recently the gradual adoption of the Rust programming language in the kernel.
In fact, Torvalds' embrace of Rust provides a useful parallel. Some kernel developers initially resisted Rust, arguing that C was sufficient and that introducing a new language would create fragmentation. Torvalds listened to the debate but ultimately backed the Rust for Linux initiative, pushing for incremental integration. A similar pattern is now visible with AI: initial skepticism, a period of experimentation, and then active approval once the tool proved its worth.
How AI is actually used in the kernel today
AI is used in several ways in Linux kernel development today. One of the most important is automated backporting of security fixes to older stable kernels. Tools trained on thousands of patches can now identify equivalent code patterns across different kernel versions and generate backports with high accuracy. This saves maintainers enormous amounts of manual work. Another area is bug analysis: when a user reports a kernel crash with a stack trace, an AI system can search the codebase for likely causes and suggest fixes, sometimes even generating a patch that a human maintainer can review and apply.
AI is also employed in code review. Systems like the Kernel Patch Review Assistant (an internal tool used by some maintainers) highlight potential issues such as memory safety violations, race conditions, or style violations before a patch is even submitted to the mailing list. This reduces the back-and-forth that typically characterizes the review process. The net result, according to several contributors, is that the overall throughput of patches has increased without a corresponding drop in quality.
Nevertheless, the use of AI has not been without controversy. Some developers worry about the opacity of machine-generated contributions. If a patch is written by an AI, who is legally or ethically responsible for its correctness? The kernel community's answer, as articulated by senior maintainer Theodore 'Ted' Ts'o, is pragmatic: if someone objects to the use of AI at any stage of the pipeline, they must accept that their objection does not outweigh the maintainer's right to choose their tools. Ts'o wrote: 'I don't think it's obvious that we must bend over backward to oblige the needs of all patch authors.'
James Bottomley, another top kernel developer, added: 'The contributor doesn't get to approve the tools the maintainer uses to assess and apply patches. If there's AI in there, and the contributor is an AI luddite, then the patch doesn't get applied (i.e., your right to ignore AI stops when it infringes others' right to use it).'
The Software Freedom Conservancy controversy
The immediate trigger for Torvalds' remarks was the SFC's policy statement 'When Using LLM-backed Generative AI Systems for FOSS Contributions.' The SFC suggested that projects should 'support, not just tolerate, those who outright reject LLM-gen-AI systems.' This recommendation raised practical questions about how far such tolerance should extend. For instance, if a maintainer uses an AI tool to backport a patch, and a contributor objects on principle, should the project accommodate that objection? Torvalds' answer was an unequivocal no, and he warned that the SFC's advice could lead to an unworkable development environment.
The SFC later clarified that its statement was meant to encourage inclusive practices, not to dictate policy. However, the exchange highlighted a growing divide in the open-source community. Some projects, like Zig, have indeed banned AI-generated code outright, citing concerns about copyright, quality, and community norms. Others, like the Linux kernel, have taken a permissive stance, leaving the choice of tools to individual maintainers. This diversity reflects the broader ecosystem's decentralized nature, but it also means that developers who are uncomfortable with AI may find themselves increasingly at odds with the mainstream of software development.
The bigger picture: AI in software engineering
The Linux kernel's embrace of AI is part of a larger trend. In the past two years, AI-assisted coding tools have moved from experimental novelties to essential productivity aids for many developers. GitHub Copilot, Amazon CodeWhisperer, and similar systems have become commonplace in enterprises and open-source projects alike. The quality of code generation has improved markedly, especially for repetitive tasks, boilerplate, and pattern matching. In 2026, models like Anthropic's Claude Opus 4.8 and OpenAI's GPT-5 have achieved levels of reliability that make them credible collaborators for experienced programmers.
At the same time, concerns remain about copyright and licensing of AI-generated code. Some lawsuits have been filed over training data, and the legal landscape is still evolving. The kernel community's approach has been to require that all contributors, whether human or AI-assisted, certify that their work complies with the project's licensing requirements. This is enforced through the Developer Certificate of Origin (DCO) process, which every patch must pass. Maintainers have noted that AI-generated code that does not add a proper DCO is simply rejected, regardless of its quality.
Torvalds' final remarks on the topic were characteristically blunt. 'In the kernel community, we do open source because it results in better technology, not because of religious reasons. And so we make decisions primarily based on technical merit. Not fear of new tools.' That principle has guided Linux for more than three decades, and it remains the foundation upon which the project's future will be built. For now, at least, AI has earned a place in that foundation.
Source: ZDNET News