Linux AI‑code Policy

The Linux kernel project now says developers may use artificial intelligence tools to write code, but the human who submits a patch remains fully responsible for every bug, license issue, and review comment. (docs.kernel.org) The policy appears in two new documentation pages, “AI Coding Assistants” and “Kernel Guidelines for Tool-Generated Content,” both now published in the kernel’s official docs. The rules say artificial intelligence agents cannot add the legally binding “Signed-off-by” line, because only a person can certify the Developer Certificate of Origin. (docs.kernel.org 1) (docs.kernel.org 2) Contributors who use a model for a “meaningful amount” of a patch must disclose it with an “Assisted-by” tag that names the agent and model version, and the docs say they should describe prompts or summarize longer sessions in the changelog or cover letter. The guidance also says basic tools such as Git, GNU Compiler Collection, Make, and editors do not need to be listed. (docs.kernel.org 1) (docs.kernel.org 2) Linux is the core software layer that talks to hardware and underpins Android, servers, and most cloud systems, so kernel patches are reviewed line by line before they are merged. The new documents keep that process in place and frame artificial intelligence as another code-generation tool, not as a new source of authority. (kernel.org) (docs.kernel.org) The language also reflects a practical constraint inside the project: reviewer time is limited, and maintainers want to know which parts of a submission came from a tool. The “tool-generated content” page says that transparency helps preserve trust between submitters and reviewers while keeping development healthy. (docs.kernel.org) The kernel community has dealt with trust problems before. In April 2021, Linux maintainers froze contributions from the University of Minnesota after researchers submitted intentionally buggy patches as part of a study, according to LWN’s contemporaneous report. (lwn.net) The current artificial intelligence rules were finalized after months of debate inside the kernel world, according to reporting published on April 13, 2026, by ZDNET. That report said the disclosure tag gained momentum after Nvidia engineer and kernel developer Sasha Levin submitted an artificial-intelligence-generated patch without initially telling reviewers the code, changelog, and tests had been machine-written. (zdnet.com) The result is a narrow rule, not a blanket endorsement. Linux accepts artificial intelligence assistance the same way it accepts any other tool: show your work, follow the existing legal process, and put your own name on the consequences. (docs.kernel.org 1) (docs.kernel.org 2)