Linux kernel sets pragmatic rules for AI-assisted contributions

What surfaced on Hacker News

A Hacker News post from 2026-04-10 brought attention to a new Linux kernel document called AI Coding Assistants. The thread had 311 points and 205 comments when reviewed, which made it one of the clearer signs that open-source infrastructure projects are moving from abstract AI debates to concrete contribution rules. The linked text is not a personal blog post. It lives in the main kernel tree as project process documentation.

The guidance is notable because it does not try to ban AI tools outright. Instead, it says AI-assisted work must still follow the normal kernel workflow, licensing rules, coding style, and patch submission process. The practical center of gravity is responsibility. The document says AI agents must not add Signed-off-by tags, because only a human can certify the Developer Certificate of Origin. It also says the human submitter is responsible for reviewing AI-generated code, ensuring GPL-2.0-only compatibility, and taking full responsibility for the final patch.

The document also introduces an Assisted-by attribution format. That gives maintainers a way to record which agent or model was used, plus optional specialized tools such as coccinelle or sparse, without treating ordinary build tools as AI. In other words, the kernel project is separating disclosure from authorship: tools can be acknowledged, but legal accountability stays with the person who sends the patch.

The Hacker News reaction was mostly pragmatic. Several commenters described the policy as common-sense because it allows experimentation without weakening maintainership or DCO rules. Others argued that attribution language does not magically remove infringement risk if generated code includes licensing problems. That tension matters because the document is less a final answer than a governance template. It shows how a large open-source project can permit AI assistance while still drawing a hard line around review, provenance, and legal certification.

For AI/IT watchers, the broader signal is that infrastructure projects are beginning to operationalize AI usage instead of debating it in the abstract. If similar rules spread to other large repositories, the industry may converge on a simple norm: AI can help write code, but humans must still own the patch, the license chain, and the final sign-off.

Source links: Hacker News thread, Linux kernel document.