Linux Kernel Drops Its AI Disclosure Tag: What Changed and Why
Linux kernel maintainers voted to deprecate the AI-assisted commit tag. Here's the reasoning, the pushback, and where Linus Torvalds stands.
What happened to the Linux kernel’s AI disclosure tag?
At the Linux kernel Maintainer Summit, a gathering of top kernel maintainers, the group reached consensus to deprecate the “assisted-by” tag, a commit line that flagged when AI had been used to help write code or a patch. The documentation requiring the tag is being removed. Sasha Levin, who has been central to the kernel’s various AI policy discussions, announced the decision. A replacement mechanism serving a similar purpose is being discussed, but as of now the formal requirement is gone.
TL;DR
- The Maintainer Summit reached consensus to deprecate the assisted-by tag and drop the documentation that required contributors to disclose AI use in commits.
- Maintainer Laurent Pinchart argued the tag was still useful as a review tool, since asking submitters to add it (and watching them go silent) exposed undisclosed AI-generated patches.
- Greg Kroah-Hartman previously pushed for disclosure not because he’s anti-AI but because he wants a signal he can review against, and he’s acknowledged the policy may need more documentation around it.
- Critics called the tag unenforceable, pointing out that AI-written commit messages are already hard to distinguish from verbose human ones, and the signal is getting weaker as models improve.
- Linus Torvalds restated his long-standing position that the kernel is not an anti-AI project and compared banning “AI-assisted” contributions to banning compilers or text editors for not proving the author’s competence.
- Torvalds has gone from cautiously optimistic in 2023 to actively using an LLM coding assistant (Claude Code) himself, crediting it with code, tests, and even commit text under his direction.
- A possible middle ground floated in the discussion is moving disclosure out of the Git commit log and into a cover letter instead, which several participants said would solve the same problem without permanently baking AI metadata into kernel history.
Other agents ship a demo. Remy ships an app.
Real backend. Real database. Real auth. Real plumbing. Remy has it all.
Why did maintainers want the tag in the first place?
The assisted-by tag existed to create a record: was AI involved in generating or supporting a given commit, and if so, how? Proponents saw at least two uses for this. First, data collection: tracking how much AI is used in the kernel over time, which developers rely on it, and how that changes as tools improve. Second, and more practically, it gave reviewers a lever. Laurent Pinchart made the case during the summit that the absence of the tag was itself useful information. If a patch looked AI-generated and a maintainer asked the submitter to confirm or add the tag, submitters who were cutting corners would often just stop responding. That silence was a signal in itself, one that disappears if there’s no formal tag to ask about.
Greg Kroah-Hartman’s position, cited in the kernel mailing list discussion, was similarly pragmatic rather than hostile to AI: “All I want is the signal that someone is using the LLM so that I can review it as such. And if I think someone is not reporting that, I can ask them to properly attribute it, and if they lie, well, that’s on them.” He later said he’d shifted his thinking somewhat and that better documentation around disclosure is probably still worth having.
Why did maintainers vote to remove it?
The core objection was enforceability. Kees Cook noted that his own commit logs are now a blend of his own writing and LLM output, edited and reorganized repeatedly, to the point where the line between “AI-assisted” and “human-written” is blurry even to him. He pointed out that he’d already mistaken a colleague’s detailed, human-written commit message for AI output, a kind of reverse Turing test failure that suggests the distinguishing signal is disappearing.
There was also a procedural complaint: the decision to deprecate the tag was made in person at the summit, but the reasoning wasn’t documented anywhere accessible. One contributor flagged that the change reversed a previous consensus without a recorded rationale, meaning anyone who later wonders why the tag was dropped has no official record to point to, only secondhand accounts of a hallway conversation.
Pinchart also raised a pointed challenge to the “it slows us down” argument. He noted that contributors frequently describe LLMs as transformative, citing an example from the summit of Linus Torvalds doing in a couple of months what might otherwise have taken years. If the tooling genuinely delivers that kind of advantage, Pinchart argued, then adding a short disclosure line to a commit message is a negligible extra cost, not a real burden.
What’s the proposed alternative to the commit tag?
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
Rather than embedding disclosure in the Git commit log itself, where it becomes a permanent part of kernel history, Pinchart suggested moving it to the cover letter of a patch series, or placing it below the “scissors line” used to mark non-committed content in an email submission. This would preserve the signal for reviewers at submission time without requiring every individual commit to carry AI metadata forever. Pinchart said this was his preferred approach, though he noted it didn’t get strong support at the summit. The debate over where (or whether) to formally capture this information appears unresolved, even though the original tag requirement has been dropped.
Where does Linus Torvalds actually stand on AI in the kernel?
Torvalds has been explicit, more than once, that the kernel is not and will not become an anti-AI project. Back in July, he stated plainly that Linux “is not one of those anti-AI projects” and that anyone with an issue with that is free to fork the project or walk away. He’s framed AI as a tool, not fundamentally different in kind from other tools developers already rely on, and said its usefulness is “no longer in question.”
That’s a shift from his position in 2023, shortly after ChatGPT’s release, when he was more measured. Asked then whether LLM-written code would end up submitted to him, Torvalds said he was convinced it would happen eventually, possibly already on a small scale, but he wasn’t particularly worried about it. He framed it as another step in a long history of automation helping people write code, and expressed hope that AI tools could catch “obvious stupid bugs” that don’t require deep intelligence to spot. When asked about hallucinations, he was characteristically blunt: “I see the bugs that happen with our AI every day, so that’s why I’m not so worried. I think we’re doing just fine and making mistakes on our own.”
By 2025, Torvalds had moved from commentary to practice. He committed code to the tools section of the kernel where he disclosed that the test programs, review, and first draft of the accompanying text were produced with an LLM coding assistant (Claude Code) under his direction, with measurements run on his own machine. It’s a rare case of Torvalds being directly involved in AI-assisted code submission rather than just weighing in on the policy.
In the assisted-by tag thread itself, Torvalds dismissed the idea of banning AI-assisted contributions outright, comparing it sarcastically to banning compilers or text editors on the grounds that using them hides whether a contributor “really understands” the underlying machine. His stated expectation for contributors is that they understand what they’re doing and do their best, and he argues that has very little to do with which tools they use to get there.
Is the kernel becoming more or less strict about AI?
Less strict, in terms of formal disclosure requirements, but not indifferent. The removal of the assisted-by tag doesn’t mean maintainers stopped caring whether AI was involved in a patch. It means the mechanism for tracking it shifted from a mandatory commit field toward informal reviewer judgment and ongoing discussion about better alternatives, like disclosure in cover letters. Individual maintainers like Kroah-Hartman and Pinchart still want signal into AI use for review purposes. The difference is that the kernel community concluded a required tag wasn’t delivering reliable signal anyway, since dishonest submitters could simply omit it and honest ones were already producing commit messages indistinguishable from AI output.
Frequently Asked Questions
What was the assisted-by tag in the Linux kernel?
Remy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
It was a line added to Git commit messages disclosing that AI had been used in some capacity, whether generating code, assisting with a commit message, or supporting review, with documentation that required contributors to include it.
Why did Linux kernel maintainers remove the AI disclosure requirement?
Maintainers at the Maintainer Summit concluded the tag was largely unenforceable. Dishonest submitters could skip it, honest contributors’ AI-assisted commit messages were becoming hard to distinguish from human-written ones, and some argued a replacement mechanism (like disclosure in a cover letter) would serve the same purpose better.
Does this mean the Linux kernel doesn’t care about AI-generated code?
No. Maintainers including Greg Kroah-Hartman and Laurent Pinchart still want visibility into AI use for review purposes. The change removes one specific, flawed enforcement mechanism, not the broader interest in knowing when AI contributed to a patch.
What is Linus Torvalds’ position on AI in kernel development?
Torvalds has said repeatedly that Linux is not an anti-AI project and considers AI a tool like any other. He has used an LLM coding assistant himself for tests, code, and commit text, and argues that restricting AI-assisted contributions is as arbitrary as banning compilers or editors.
Is there a replacement for the assisted-by tag?
A formal replacement hasn’t been finalized. One proposal discussed at the summit was moving AI disclosure out of the commit log and into the patch cover letter instead, but it didn’t have strong consensus at the time of the discussion.
