Linux kernel developers move to drop the rule that made coders disclose AI help

Every change to the Linux kernel arrives as a patch with a commit message. That message carries short tags, like who wrote it and who reviewed it.

The Assisted-by tag was a newer one. The policy said anyone who used "any sort of advanced coding tool in the creation of [their] patch" should add it, The Register reports.

The kernel's guide for AI coding assistants shows how it looks. It names the model and, optionally, specialized analysis tools such as coccinelle or sparse. Basic tools like git, gcc, make and editors were never meant to be listed.

• July 2025: the labeling proposal first took shape, The Register reports.
• January 2026: it became part of the kernel's official documentation.
• July 2026: developers argued about simplifying it or dropping it, Phoronix reported at the time.
• October 2026: the Maintainers Summit agreed to deprecate it.

The Linux Maintainers Summit met in Prague last week, Phoronix reports. That is the yearly meeting where the people who look after the big parts of the kernel sort out how the project runs.

There was consensus to drop the tag. "The Maintainers Summit reached consensus to deprecate the Assisted-by Tag," Sasha Levin, a kernel developer and engineer at Nvidia, wrote on the kernel mailing list, according to The Register. "Remove the documentation requiring it."

Levin's patch does exactly that: it removes the documentation that lays out the tag. Phoronix says the mailing list thread shows mixed views. But because key maintainers already agreed at the summit, it expects the change to go through in the near future.

In real life If you send a fix to the Linux kernel next year after an AI assistant helped you find the bug, you will most likely not have to add a line saying so. You will still have to sign the patch yourself and stand behind every line of it.

The arguments started in the summer. Christian Brauner, a prominent kernel developer, opened a thread in July asking to simplify the rule, Phoronix reported then.

"I find it very irritating that our git history has effectively started to function a bit like a free advertising platform for a bunch of AI companies and their proprietary agents and models," Brauner wrote.

A day later, Red Hat developer Jeff Layton proposed a patch to remove the requirement entirely. He gave three reasons:

• Not everyone follows it: many people attribute AI tools in good faith, but not all do.
• Free advertising: the tag "basically serves as free advertising for proprietary LLM companies."
• No clear purpose: "it's not clear why we want to collect this info in the first place."

Layton summed up that the data was "flawed at best" and collected for a purpose that was not clear. Phoronix added another complaint from the thread: the tag never said how the AI was used, whether it wrote the patch or only helped with testing or review.

The big picture: the tools got better. The Register notes that AI disclosure was not an issue while generated code was obvious, buggy and bad. In the past year or so, coding agents started producing code that meets the kernel community's bar.

"LLMs are good enough now," veteran maintainer Greg Kroah-Hartman said at the Linux Foundation's Open Source Summit in South Korea in August, according to The Register. He said AI has gotten good enough to find bugs because "code is nothing but patterns in a specific format."

Not everyone is on board. Laurent Pinchart, a kernel developer and founder of Ideas on Board, objected on the mailing list on Tuesday.

"I'm not happy removing the tag without adding another mechanism for required disclosure," he wrote, according to The Register. "Without that, I can foresee submitters pushing back on the question."

Pinchart also challenged the idea that a rule some people ignore is useless. He argued it is about setting expectations: saying what the project wants and what it does not. People who cheat have always existed, he wrote, and their contributions get rejected.

Back in July, another reviewer made a related point on the earlier patch. A separate kernel document on tool-generated content still asks contributors to disclose code-generation tools, so dropping the tag alone could leave the rules out of step.

Torvalds answered Pinchart directly, and he did not hedge. He said he does not understand the idea of "don't use AI for X."

His argument used older tools to make the point. He suggested, tongue firmly in cheek, a rule against using compilers, so contributors would have to prove they understand the machine by writing everything in assembly. Then he added a ban on text editors for commit messages, so nobody could hide their typos with the backspace key.

"So no. We're not going to make stupid rules that say 'you can't use tools for X'," Torvalds wrote, as quoted by The Register.

==His actual expectation is simple: "The kind of expectations I have is that people do their best and I expect people to understand what they are doing."== That understanding, he wrote, has very little to do with which tools they use.

He closed by saying the people he respects "are not the ones that shy away from using new and interesting tools." The Register notes he has warmed to AI tools over the past year, after early skepticism about how vibe coding affects maintainability.

Dropping the tag does not mean anything goes. The kernel's AI guide puts responsibility squarely on humans:

• No AI sign-off: AI agents must not add Signed-off-by tags. Only a human can legally certify the Developer Certificate of Origin, the kernel's promise that you have the right to submit the code.
• Human review: the person submitting must review all AI-generated code.
• Licensing: every contribution must be compatible with GPL-2.0-only.
• Full responsibility: the human submitter takes full responsibility for the contribution.

The catch: worries have not gone away. The Register lists errors slipping through unnoticed, accountability, and coding agents reproducing copyrighted code. Open-source projects are also dealing with floods of low-quality AI reports, which is why Google paused part of its open-source bug bounty this month.

Even the AI companies disagree on labels. The Register notes that Claude adds provenance messages to code by default, while OpenAI's Codex does not.

• If you contribute to Linux: you will likely not need an Assisted-by line once the patch lands, but reviewers still expect you to explain and defend every line.
• If you use AI to code at work: the kernel's stance is a strong signal that "understand what you ship" is becoming the rule, not "say which tool you used."
• If you run an open-source project: expect this debate in your own repo. Pinchart's point about setting expectations is worth reading before you copy the kernel's choice.