https://docs.kernel.org/process/coding-assistants.html
Broadly it says: you can use AI as much as you like. You have to disclose use of AI. You, as a person, not the AI, are responsible for the quality of Linux kernel submissions.
As someone who doesn't know much about linux kernel work: what does being responsible for the submission look like in practice? Are you expected to find fixes if your change causes unanticipated issues?
It's just another way of saying, you can't have AI submit things on its own and you can't "blame" AI for anything you submit, you're expected to understand and be able to explain anything you're putting forward, if there is an issue with your submission, doesn't matter at all if you go back to AI to address it, so long as all issues are properly addressed.
The code is either of sufficient quality, or it's not, doesn't matter whether you or AI wrote it, it's your name on it, you're responsible for it and its your reputation on the line.
Should be applied to anything someone is releasing without big disclosures.
At least until a few months from now when it will be clear that human code is the weak point.
(And in reality EVERYONE's AI policy is a variation of this no matter how strict they get, because very few people will suddenly rip all of tridge's code out of Samba if he admits he used AI once, and nobody can withstand slopspam for long.)
At least that's how I do it. When someone offers a patch to an open source project, I don't really care if that person has a PhD, or if that person does this in their free time, or if that patch is actually the result of a whole team working together. I judge the patch based on the value it brings to me as a maintainer.
Obviously the code has to be good, but also the interaction with whoever offers the patch has to be smooth: if I have a question, if I need them to confirm/change/test/elaborate on something, I need that to work.
If I feel like I would be faster doing the work myself AND this author is not putting effort into it (e.g. they just copy-paste an AI), I will not review that work. Possibly I will distrust that author. If too many contributions end up wasting my time, I will start limiting the contributions to people I know/trust, etc.
In terms of maintenance, I don't see a difference between getting a contribution from a stranger or from a stranger using an AI.
In terms of philosophy, that's obviously different: One may not want to accept contributions from people using AI because they think that AI is bad for society, for instance. And I would totally understand that. But if the stance (and it seems to be Linus' stance) is that a patch is only judged based on its technical merit, then I don't see why AI-in-the-loop would be an issue.
Perhaps not in terms of maintenance. In terms of provenance there is a significant difference between "I wrote this." and "I found this."
Which means that "as a maintainer", I don't write anything. I merge code that falls from the sky. Whether it was an AI cloud or a manual-coding cloud, I don't know: from where I stand, it falls from the sky.
> there is a significant difference between "I wrote this." and "I found this."
Of course. When I write code ("I wrote this"), then it does not fall from the sky: I wrote it. I know, it's me :-). But in the open source world, being a maintainer means that you deal with code that falls from the sky. Unless you decide to close your project to contributions and only accept people you trust.
That's very well put. I'm stealing your wording.
The thing which I think is going to be a very sore lesson for everyone at some point is that if anyone uses AI it becomes a burden for everyone. Even in projects that reject all AI usage.
What I can do is create 50 personas that all contribute patches. 40 of them get called out and are distrusted, but 10 make it through. I can give more focus on those 10 to ensure their trust level increases. Then slowly have them all come together to shepard some subsystem where I'm going to be the only real person involved and can sign off on my own work in a that looks like I have oversight from the outside.
In the past this was possible but took a very high cost. You need a bunch of competent software engineers who you can trust not to blab their secret mission. Today anyone can give it a try.
However, I remember that a couple of university researchers (PhD candidates maybe?) thought it would be a good idea to introduce security issues in Linux, and managed to do it. Without AI and without a lot of resources.
A consequence of that was that Linux immediately reverted all the patches coming from that university, saying "we don't know if we can trust you guys" (but really I guess as a way to tell universities "if you fuck with us we ban you, now keep your people in line").
And that's what I was thinking about when I said that if too many contributors misbehave, as a maintainer I will just stop accepting contributions from people I don't know. And maybe that will happen. Maybe it will be harder and harder to just be accepted into the club of the people who can send a contribution and have it reviewed at all.
That last part is worth highlighting: you always have the option to not follow-up and abandon your proposed change, in which case it won't make it into the kernel. (Unless someone else picks it up and volunteers to get it into shape for that is acceptable for merging into the kernel, but that is vanishingly rare, unless the change is trivial.)
You aren't expected to hang around and maintain your patch once merged, that's what subsystem maintainers are for. But you would be making their lives a lot easier and earn a lot of respect if you did.
How long until we get "all FOSS code is made by AI and so it's not copyrighted" arguments when using code in unlicensed ways?
It also hides what would be a great signal on which ML models to use to get the best quality code. Like, wouldn't the community like to know which models made the code with the least bugs?
(This is why fan made videos using copyrighted characters can be taken down.)
They dont think the juice is worth the squeeze. LLMs are just another tool.
You should read reasoning behind the decision.
https://www.theregister.com/ai-and-ml/2026/10/14/linux-kerne...
[0] https://lwn.net/Articles/1099594/ (user required).
[1] https://lore.kernel.org/lkml/20261009083137.3-1-sashal@kerne...
If you spent 20 seconds to generate a patch and just hurled over the fence to be reviewed is no worse than spending 3 hours to manually write a fix that fixes 1 thing and break 2 others.
Nothing says you have to use AI. Just like nothing says you have to use a computer. You could have someone print the code for you, write your patch with a pencil on paper, get a pigeon to send your patch to someone who can then submit it, and it would be judge in the same manner. Granted, it's probably more efficient to use a computer.
Except your employer, usually.
You are right, that many of the more serious, less casual contributors are probably paid.
This is no longer true.
*Linus does it*
Linux maintainers: Goddammit!