As AI programming tools such as GitHub Copilot, Anthropic Claude, and ChatGPT Codex become increasingly popular in development environments, large organizations have also begun to adopt AI-assisted code on a large scale in enterprise-level software. For example, Microsoft internally claims that AI Copilot has been involved in reviewing and influencing hundreds of thousands of pull requests every month. The community now also has clearer written guidelines for how AI-generated or assisted-generated code can be used compliantly in the Linux kernel.
Recently, a document on AI-assisted code requirements has attracted attention on Hacker News. This document is publicly hosted in the official Linux repository under the name of Linus Torvalds in the form of RST, specifically to provide guidance and constraints for developers who want to use AI tools to contribute code to the Linux kernel.
According to the document, first of all, all AI-assisted generated code must fully follow the existing human development process, including the process and style requirements specified in development-process.rst, coding-style.rst, submitting-patches.rst and other files. At the same time, the code must also meet existing compliance requirements, such as remaining compatible with the GPL-2.0-only license, correctly using the SPDX license identifier, and adhering to all rules listed in license-rules.rst.
In terms of signing and responsibility identification, the rules emphasize that the AI agent itself must not use the Signed-off-by label. This label is reserved for human reviewers only, as only humans can legally certify a Developer Certificate of Origin (DCO). Human developers submitting patches must personally review all AI-generated code to ensure it meets licensing requirements, and use their own Signed-off-by mark on the patch to clearly take sole responsibility for all contributions.
The document also requires explicit attribution of AI-assisted participation so that the community can track the evolution of AI's share of Linux kernel development. This is done by using the “Assisted-by” tag in pull requests and following a consistent format:AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2]. Among them, AGENT_NAME is the name of the AI tool or framework used, MODEL_VERSION is the specific model version, and TOOL1 and TOOL2 in square brackets are optional special analysis tools, such as coccinelle, sparse, smatch, clang-tidy, etc.
Against the background that Linux kernel developers and major organizations continue to embrace AI and use it to improve coding and review efficiency, this specification maintained by Linus Torvalds is actually "conditionally welcoming" AI-assisted code into the Linux kernel: it does not deny its efficiency advantages, but also sets clear boundaries for AI participation through hard requirements such as human responsibility, license compliance, and unified attribution formats. It has been previously reported that Torvalds himself has also tried to adopt new development practices such as so-called "vibe coding" in some projects, which also confirms his pragmatic and prudent attitude towards new tools.
learn more:
https://github.com/torvalds/linux/blob/master/Documentation/process/coding-assistants.rst