Linux Kernel Drops Its Assisted-by AI Tag After Only 10% of 7.3 Commits Used It

TL;DR

The Linux kernel no longer asks contributors to tag AI-assisted patches. Jonathan Corbet's docs-7.4 pull, merged into mainline on October 21, includes Sasha Levin's patch that deletes the Assisted-by requirement from the docs and from checkpatch. The reason is in the commit message: in the 7.3 cycle through -rc7, roughly 10% of 18,533 non-merge commits carried the tag, while actual LLM use is "far more widespread." The same pull added a top-level AGENTS.md, so the merge that stopped asking humans to declare AI use also gave AI agents their own entry point to the rules.

What the merge removed

The AI Coding Assistants document was committed in January and shipped with Linux 7.0. It asked anyone whose patch was developed with an LLM to add an Assisted-by line to the commit. In August, a patch from Christian Brauner trimmed the format to a bare Assisted-by: LLM, because naming the specific model "provides free advertising to proprietary software companies while adding little or no useful information."

As of October 21, the tag is gone. The patch removes the Attribution section from coding-assistants.rst and submitting-patches.rst, strips it from the Italian, Brazilian Portuguese and Traditional Chinese translations, and deletes 11 lines of tag handling from scripts/checkpatch.pl. Thirteen developers acked or reviewed it, among them Greg Kroah-Hartman, Kees Cook and Miguel Ojeda.

Some rules stay. AI agents still must not add a Signed-off-by, since only a human can certify the Developer Certificate of Origin, and the human submitter still owns the patch. The separate tool-generated content guidelines are unchanged. They still ask you to say which tools you used, in cover letters and changelogs, and to note it in the changelog if a tool found the bug you are fixing.

A label that only one patch in ten wore

The numbers come from Corbet's LWN report on the 2026 Maintainers Summit, held October 8 in Prague. Linux 7.3 took a record 18,533 non-merge changesets through -rc7. Only 1,884 of them carried Assisted-by.

LWN calls the true number of LLM-assisted patches "surely much higher." At the Summit, Dave Airlie estimated that "probably 90%" of contributors use LLMs, and said many of them fear an "anti-LLM witch hunt." Linus Torvalds thought plenty of submitters were ashamed to admit it, and that the requirement "might scare people away from using LLMs." Corbet argued that "Assisted-by provides no useful signal."

Think of a "may contain nuts" label printed on one bar in ten when the factory puts nuts in nine out of ten. You learn nothing from the bars that don't carry it, and the label on the bars that do isn't telling you much either. Levin's commit message makes the same point in kernel prose: the tag's "absence says nothing about how a patch was made, and for developers who use these tools for almost everything, its presence says little more."

The commit frames it as a choice between enforcing the requirement and dropping it, and there was "a strong, though not unanimous, consensus in the room to drop it." Thomas Gleixner half-jokingly suggested a Not-assisted-by tag instead, and Arnd Bergmann said that would actually help him find developers to mentor. Maintainers still want to know whether a human was involved and how a patch was made. The view at the Summit was that this belongs in the cover letter, where you can describe how the tools were used, not in a one-line trailer.

Same tag, different vote three months later

This had been tried before. A proposal to drop the tag was turned down in July, mainly because maintainers wanted a signal that a patch was LLM-assisted so they could review it accordingly. Kroah-Hartman argued for keeping it then. He supports removing it now, per the commit message.

The mailing list was not unanimous. Jeff Layton of Red Hat listed three problems, The Register reported: not everyone follows the policy, the tag "basically serves as free advertising for proprietary LLM companies," and "it's not clear why we want to collect this info in the first place." Laurent Pinchart pushed back: "I'm not happy removing the tag without adding another mechanism for required disclosure." Torvalds called a "don't use AI for X" rule "UTTERLY STUPID" and wrote that "It has nothing to do with 'expectations', and it has everything to do with your hangups." His expectation, he wrote, is that "people do their best and I expect people to understand what they are doing."

Phoronix covered the plan on October 13, when it was still a patch. The October 21 merge is when it became policy.

Written for the agents now

The other AI change in the docs-7.4 pull went the opposite way. Levin's AGENTS.md commit adds a top-level AGENTS.md as a symlink to README, because most coding agents load an AGENTS.md automatically and often don't read the README before acting. The README has a section that tells any LLM it "MUST read and follow" the coding-assistants document.

The commit includes a small experiment. Levin asked two agents to "create a new commit renaming the release (in the Makefile) to AI Test." Without AGENTS.md, one agent added a Signed-off-by for the human on its own and used its own Co-Authored-By trailer, and the other added no attribution at all. With AGENTS.md, neither added a sign-off, both followed the kernel's patch guidelines more closely, and both wrote Assisted-by: LLM.

So the evidence that AGENTS.md works is two agents correctly writing a tag that the same pull request deleted. (The commit was written on September 24, two weeks before the Summit.) The README still describes the coding-assistants document as covering "licensing, attribution, and the Developer Certificate of Origin," and the attribution part has since been deleted.

What this means if you ship code with agents

• Kernel patches no longer need a trailer. If an LLM did real work, describe how you used it in the cover letter. The generated-content guidelines still ask for that, and maintainers said at the Summit that they want it.
• A coding agent pointed at a kernel tree will now load AGENTS.md on its own. The binding rule is still that no agent signs off. Its default trailer depends on the tool: The Register notes Claude adds provenance by default and OpenAI's Codex does not.
• If you copied the kernel's tag into your own CONTRIBUTING.md, look at what happened here first. The kernel ran a mandatory tag for most of a year, few maintainers asked about it, and about 10% of commits carried it. Check the tag in review or expect the same result.

The kernel's process docs now address two readers: humans, who are trusted to explain their tools in prose, and agents, who get a file at the repo root telling them what they may not do.

Key Takeaways

• The docs-7.4 merge on October 21 removed the kernel's Assisted-by requirement from the docs, their translations and checkpatch.
• Only 1,884 of 18,533 non-merge commits in Linux 7.3 through -rc7 carried the tag, about 10%, while maintainers estimated real LLM use was much higher.
• The Maintainers Summit in Prague on October 8 chose to drop the tag rather than enforce it, three months after a July proposal to drop it failed.
• The DCO rule stays: AI agents must not add Signed-off-by, and tool use belongs in the cover letter under the generated-content guidelines.
• The same pull added a top-level AGENTS.md symlink to README, so coding agents now pick up the kernel's AI rules automatically.

Sources: Linux commit 1e035297, "docs: remove Assisted-by references", Linux merge 14c89774, docs-7.4 pull, Linux commit 90d9f76c, "docs: add AGENTS.md as a symlink to README", Linux commit 816d9992, "coding-assistants: simplify attribution", LWN: The 2026 Maintainers Summit, The Register, Phoronix, Kernel docs: AI Coding Assistants, Kernel docs: Tool-Generated Content