Linux project founder Linus Torvalds has published the first kernel development patch created with the assistance of a large language model. The solution aims to address a long-standing issue of excessive cache contention that occurs when multiple processes access the same file simultaneously. According to Torvalds himself, in a synthetic benchmark, the proposed solution delivers a 23-fold performance increase, although in real-world conditions the effect will be significantly more modest. Nevertheless, as the developer clarified, at the scale of server systems where such load scenarios constantly occur, the difference promises to be tangible.
A Problem That Could Not Be Solved for Years
Cache contention during simultaneous access by multiple processes to a shared file is one of those issues that periodically resurfaces in Linux kernel discussions. The scenario is typical when working with shared libraries, databases, or web servers: dozens and hundreds of processes try to read the same memory pages, and the system wastes resources resolving conflicts. According to the source, attempts to solve the problem were made repeatedly, but none of the proposed options became universal or was accepted into the main tree. Torvalds, by his own admission, prepared his own version back in 2024 and has been using it on his personal laptop since then, but it never made it to public release. The turning point was the decision to "try to tackle this task properly with the help of AI."
A Shift in Position: From Skepticism to a Tool
Torvalds' attitude toward AI-written code has not been static: over the past few months, he has repeatedly changed the focus of his statements. In May 2026, he sharply criticized neural network-generated bug reports, which he said clogged the incoming message stream and hindered normal work on the kernel. However, by July, his stance had shifted: Torvalds announced his intention to strictly suppress attacks by AI opponents within the community, emphasizing that new development tools often handle tasks better than humans and help identify elementary errors. A key step was lifting the obligation for patch authors to report whether they used AI when writing code: such tools are now equated with standard development tools, such as compilers and text editors.
The Scale of AI's Impact on the Linux Kernel
Statistics cited in the context of the patch publication demonstrate how deeply AI has already penetrated the kernel development process. In the Linux kernel alone, about 2,000 vulnerabilities (CVEs) are eliminated with each release, and a significant portion of this work is done using automated tools. In parallel, developers have begun phasing out support for legacy hardware drivers, as it is in these that a disproportionately high number of vulnerabilities are discovered. Under these conditions, a patch created using a language model and having passed the community's rigorous review procedure looks less like an experiment and more like a logical stage in the evolution of the development process.
What This Means for the Community
The publication of the first "AI patch" in the main Linux kernel tree is a symbolic event. The Linux kernel is traditionally considered one of the most conservative projects in the open-source world: every byte of code goes through multi-stage reviews, and Torvalds himself is known for his intolerance of substandard solutions. The fact that a patch created with the participation of a neural network passed this filter and was accepted means that for the community, AI has ceased to be an "alien" tool. At the same time, Torvalds maintains his characteristic caution: he himself stipulates the 23-fold increase in the synthetic benchmark as an upper bound that does not reflect real-world operation. But for server infrastructures where cache contention is a daily norm, even a modest percentage performance increase translates into measurable resource savings and reduced latency.
