Linus Torvalds, about AI, said that if people don’t like it they can fork it. Does anyone know any linux kernel forks that attempt this? Getting rid of “tainted” commits, even if that would mean going back in time a few years?
I don’t want to discuss the use of AI in the kernel. I am just curious and couldn’t really find any forks (only the linus quote pops up on search engines).
I’ve been thinking along the same lines, actually. I’d personally love to switch to a no-AI kernel fork.
It seems someone has at least done the work to find the last commit not tainted by slop. So that’s one tiny tiny baby step in the direction of such a fork, I guess.
Honestly, I’d happily sacrifice the newest/latest features of Linux. I could just stick with version 6.16.7 (which according to the above link is slop-free), except that some fairly impactful vulnerabilities have been discovered in Linux lately.
I use Gentoo and configure/build my own kernel manually, so I could also theoretically do some research and see how many of the vulnerabilities that are actively a problem in 6.16.7 are in parts of the kernel that I explicitly disable anyway.
I’m no C developer, though, let alone Linux kernel hacker. If I had the chops, I might honestly be considering making a fork myself right now.
I’m in the way of getting it off the ground, though I still don’t know where to branch off from the main branch. First known AI slop contribution?
What should be the name? While “zilix” would be an obvious one, I want to avoid the ego issue Linus Torvalds have developed. Name it “Rockhopper” maybe? Or some other species of penguin? Leave the penguin branding in the past with Linus?
But I know that I’ll try to reach out to contributors of Linux who are dissatisfied with the current direction of the kernel. And that I will try to lobby some memory safe API into the kernel (SafePosix?), meaning null terminated strings will be the no longer preferred alternative, for backwards compatibility only, though some functions may still return a null terminated string for simplicity sake, but now alongside with a length indicator.
Would zilix not just be a reference to ziltoid the omniscient? Or is your username based on a part of your real name? Otherwise i dont see why you’d develop ego issues over a name that isn’t yours.
Technically “Linux” is also just a general reference to “Linus”, and not “Torvalds”, hence it’s not called “Torvalx”.
Sadly the issue is there is no community around any other particular fork. Technically, there are many forks, mostly for hardware / distro reasons. If you wanted to be sure there are no “AI / LLM” contributions in the kernel, 5.15 has an epoch around February. The other series but I’m not sure. I’d personally like to see a reset probably further back.
I looked into this a few months ago to see how I’d go about addressing it but the hope was that they would at least leave the ”Assisted By” tag to track these types of commits… but sadly I was wrong. Sadly, the Linux kernel and the community around it are no longer trustworthy.
If you’re concerned about it being a security vulnerablility use the Linux LTS kernel (I think version 6.9.5 or something along those lines)
I think you’re probably going to need to go back a few years if you really want an untainted kernel. Pretty sure there have been LLM assisted commits for at least two years. And some of those likely didn’t have any indication. To be safe, you probably want to go back to at least 2023 if not 2022.
Are there any indications that AI other than the festering LLM-bubble ever added content to Linux?
I don’t need answers, I’m just interested in the question.
I am not a kernel contributor, but my IDEs have local LLMs for line based predictions for quite a while already and I use that. So if the involvement of any LLM is already the mark, then I would be guilty in a lot of work I did over the last years.
Personally I draw the line at accountability. If I contribute something I take responsibility and I (try to) make sure the quality is good. I expect the same from others. I don’t care how they got their result, as long as they can defend it.
That’s also why I am not a big fan of “assisted by Claude” and similar markers. It takes away accountability. “Whoops sorry, the agent did that”. No, you let the agent do that and didn’t care enough to validate or fix it.
As someone who is pretty anti AI in general, this is my personal opinion on FOSS projects. I fully understand and support people forking FOSS projects that have an AI policy they disagree with, but the community is so small and overworked as is I worry about further fracturing. I’m not a kernel contributor, so as long as they are keeping the same or higher standards, I’m not going to complain about where the code came from. I think your comment about accountability is really important and something I agree with as well.
And that’s why a nuanced view to define what kinds of AI contribution are done with direct human involvement, oversight, and review to create high-quality code (yes, it is possible) and what kinds are problematic and dangerous is important. Both in defense of the Linux kernel project’s position, and in criticism of it. Otherwise I feel like everyone on both sides is just going to be shouting past each other with nobody actually ever reaching the point of how we are realistically ever going to build reliable and maintainable software going forward in a world where the AI pandora’s box has already been opened and and short of an absolute total war against the technology (which may be justified, but is not likely to happen and is not achievable or winnable from the efforts of a single project) it has been released into the wild and is not going away and people are not going to stop trying to use it.
Public shaming of AI usage is a valid strategy against its adoption and possibly the most viable one we have, but plenty of people are completely immune to it and will continue to be. It’s never going to stop it completely and maybe it’s a wasted effort. At the end of the day, the work still needs to get done, and arguing about how exactly it gets done simply isn’t getting any of the necessary work done.
Well, I guess it depends on your definition of AI here. Would a linear regression fitting a curve to inform decisions on allocation or scheduling count? I wonder if anyone has also tried to apply classifiers to commits.
Sources say NetBSD is the way if your hardware supports it
Which is “better” - NetBSD or FreeBSD?
OpenBSD?
FreeBSD is more compatible. I’ve used it a bit. I like it. I wouldn’t drop Linux for it however. I’m sure people have and are currently sneaking AI code into NetBSD and other projects. If you want an AI-free kernel, you’ll need a time machine as much as that sucks. I’m not saying don’t try out NetBSD. I wouldn’t switch to it though. If you do though, more power to you. I wish FreeBSD had more of the software I use on Linux. I also wish my wi-fi card worked with it.
There are projects like RedoxOS that are slop-free, but as it is not that mature, it’s quite a bit of work to set up. It’s also implemented in Rust which is not as bad as some other projects with it, but still not free of slop.
I’m not an expert, but here is the context: https://www.theregister.com/os-platforms/2026/10/13/netbsd-may-have-found-a-new-role-the-ai-free-foss-os/5302818
You might find this useful: https://codeberg.org/ethical-foss/open-slopware#operating-systems
TL;DR - There aren’t a ton of options, especially that are daily drivable on the desktop for most people.
There is this list as well but it doesn’t seem to consider the kernel, only the distro’s stance.
