Linux 7.0 Lays Down the Law on AI‑Assisted Code

The release of Linux 7.0 this week brought more than the usual assortment of driver fixes and architecture improvements. Buried alongside Torvalds’ characteristically understated release notes was a milestone that the open-source community had been debating for well over a year: the Linux kernel now has an official, written policy on the use of AI coding assistants — and its core message is unambiguous. If you submit AI-assisted code, you own it, completely.

The new document, titled AI Coding Assistants, lives at Documentation/process/coding-assistants.rst in the mainline kernel repository. It was merged alongside the 7.0 release and sits in the same process documentation tree as the kernel’s long-established coding style guide and patch-submission guidelines.

“I suspect it’s a lot of AI tool use that will keep finding corner cases for us for a while, so this may be the ‘new normal’ at least for a while. Only time will tell.”

— Linus Torvalds, Linux Kernel Mailing List, April 2026

How the Policy Came to Be

The road to a formal policy was anything but smooth. At the 2025 Maintainers Summit, kernel security maintainer Sasha Levin pushed hard for documented consensus after a growing flood of AI-assisted patches began straining reviewers’ patience. Three principles emerged from those discussions — human accountability is non-negotiable, purely machine-generated submissions without human review are not welcome, and tool use must be disclosed — but nothing was written down at the time.

Months of internal tension followed, including a public clash between Intel’s Dave Hansen and Oracle’s Lorenzo Stoakes in January 2026 over how aggressively the kernel should police AI contributions. The frustration wasn’t simply about AI; it was about honesty. As one senior maintainer put it, developers were less angry about the tools themselves and more furious about contributors who failed to disclose their use of them — a pattern already dubbed “AI slop” by maintainers of projects like the Godot game engine.

Levin eventually committed to writing the policy and has now delivered. The result is a pragmatic, legally grounded document that neither bans AI tools nor gives them free rein.

The Three Core Rules

The prohibition on Signed-off-by from AI agents is the policy’s sharpest edge. In the Linux kernel, every patch requires a developer’s signed-off tag because it constitutes legal certification under the DCO — a binding statement that the code is GPL-2.0 compatible and that the submitter has the right to contribute it. AI systems lack the legal standing to make that certification, and the new policy makes this explicit in the document’s strongest language: AI agents MUST NOT add Signed-off-by tags.

The choice of a new Assisted-by tag rather than the existing Co-developed-by was deliberate. The Co-developed-by tag has a standing rule requiring a corresponding Signed-off-by — which directly contradicts the new AI policy. By framing AI contributions as “assistance” rather than “co-development,” the kernel community is making explicit that AI is not an author.

Linux 7.0 Itself: More Than Just Policy

The policy announcement should not overshadow the 7.0 release itself, which Torvalds noted is a version-numbering housekeeping matter as much as anything else — after 6.19, incrementing to 7.0 simply avoids unwieldy numbering. The release promotes Rust support to stable status, introduces performance optimizations, adds early support for AMD Zen 6 and Intel Nova Lake processors, and includes the usual sweep of networking, architecture, and driver fixes.

Interestingly, Torvalds’ own release notes hint at how AI is already reshaping kernel maintenance. He observed that AI tools are being actively used to surface corner-case bugs — an observation backed up by kernel maintainer Greg Kroah-Hartman, who has been running AI-assisted fuzzing in his own tree on a branch informally called “clanker,” targeting the ksmbd and SMB code. The workflow the policy is built around — AI surfaces potential issues, a human with deep kernel experience decides what is real, writes the fix, and takes responsibility — is already in practice.

What It Means in Practice

For the typical developer using GitHub Copilot, Anthropic Claude, or another LLM to help write kernel patches, the practical impact is straightforward: keep using your tools, but disclose them and review every line. The policy treats AI coding assistants the same way it would treat a compiler, a static analyzer, or a knowledgeable colleague — a tool that can help, but whose output you are fully responsible for.

Maintainers will not rely on AI detection tools to police undisclosed use. Enforcement will depend on the traditional kernel mechanisms: experienced eyes, pattern recognition, and rigorous code review. The document also reiterates that AI-generated code must conform to the kernel’s strict coding style, and that “the AI made it that way” is not an acceptable explanation when a reviewer flags a style violation.

“If the code is good, it’s good. The bottom line is that trying to ban AI tools is like trying to ban a specific brand of keyboard.”

— Summarizing the maintainer consensus, Tom’s Hardware, April 2026

A Template for Open Source?

The Linux kernel’s approach — pragmatic acceptance with strict human accountability and mandatory disclosure — may well become a reference model for other major open-source projects still working out how to handle the AI wave. The kernel’s combination of clear legal grounding, a minimal new tag, and a refusal to over-engineer the solution reflects the community’s long institutional experience with adversarial contributors, license compliance, and the costs of cutting corners in code that runs on everything from smartphones to supercomputers.

Whether the “new normal” Torvalds describes — a sustained flow of AI-surfaced bug reports and corner-case fixes — ultimately improves kernel quality or creates new review burdens remains to be seen. The policy at least ensures that whatever comes through the door, a human has staked their reputation on it.

The full policy document is available in the mainline Linux repository at Documentation/process/coding-assistants.rst and on the official kernel documentation site at docs.kernel.org.