20 November 2026
Article published by: Lhehaa Rhoeow
Date of publication: 20 November 2026
Think AI makes you super hacker? Think again! (while you still can) - see: https://arxiv.org/abs/2507.09089
Please also read this article: https://blog.codeberg.org/protecting-our-floss-commons-from-llms.html
Libreboot will make efforts, where feasible, to reduce the influence of AI on code that it uses; this may include reverting certain patches upstream, rewriting some, perhaps where feasible forking projects.
This is not a policy set in stone, because Libreboot is only a small project. There’s only so much that Libreboot can do. We cannot fight the whole world. The sad reality today is that AI is everywhere - and if it were implemented soundly, in a distributed peer-to-peer model with simultaneous investment in decent enough renewable energy infrastructure, with fully democratic and open source governance, it might not be as bad as it is today. The reality today is that a few small corporations are monopolising something that has never been monopolised before: intellect itself.
A most recent example of where Libreboot de-slopped itself was this patch: https://codeberg.org/libreboot/lbmk/commit/0ff23c0f520c9f98aad243f3304a46af0742753b - an otherwise good patch fixing a bug was optional and thus reverted. Now battery charge thresholds can’t be set automatically; they can still be set using ectool. Upstream merged thinkpad_acpi support but broke WWAN/Bluetooth, and the fix for that was AI-generated. I decided that I would just remove thinkpad_acpi support for now, as doing so would be non-destructive. For a little project like Libreboot, little patches like this are probably the best I can do, and I can reliably assure readers that there will be many more examples such as this.
This page is not an anti-AI policy insofar as it does not outright ban the use of or facilitation of AI in the Libreboot project. I know that my own personal instinct is to resist AI. I personally think AI is the worst thing we’ve ever built, and I would like it to be destroyed worldwide. This page simply clarifies the nuances of my views on the subject, and how they pertain to Libreboot development. My instinct is to delete AI, but I’m not going to be dogmatic about it either like in some other projects. That’s stupid. Let’s get first to the meat of the article before we proceed. The central idea that I wish to communicate now, is this:
Libreboot proposes a name for current iterations of generative AI: Weapons of Mass Intellectual Destruction. Why do I say that?
Simple intuition is all you really need. Remember when you were at school and there was that one kid in class who always cheated. He might get good grades but he’s really stupid and when he enters adulthood, he will be a loser because he doesn’t know how to do anything. The book nerd who studies hard is the one that will get success. AI is turning us all into cheats. It’s all too easy nowadays for people to use AI for everything, even to perform basic mental tasks that we use to do ourselves. Now, this is hypothetical, but we see evidence of it too: academic performance worldwide is suffering as a result of AI use in schools. This, combined with a reduction in the developed world of those with adult-level literacy which was already underway since the 2000s and we begin to see the problem. Don’t take my word for it. Look it up yourself.
Consider that the infrastructure for AI is economically unsustainable; when the bubble bursts, and it will, we will suddenly be left with millions of people thoroughly deskilled who will become completely useless. They will need to be re-educated, but will we even be able to afford it? In this doomsday scenario, we’ve also tied many of our economic systems and the very things that used to make us productive into AI - and indeed today, we see AI creating mass unemployment in job markets. The collapse of the AI bubble will quite plausibly lead to the world economic depression in human history.
Consider that those few companies who currently control how AI models are trained have unprecedented access to information, and control of how that information is presented; they could destroy the world if they wanted to, and subvert entire nations. They have this same power over us in the Free Software movement, as more of us begin using AI today for our software projects. We could argue that people still have the right to choose, but your right to choose is based on the very education and information you receive; fascists always attack education first, and AI is the most powerful weapon they’ve ever had.
Libreboot will do what it can to fight AI, but we use upstream projects that now openly embrace AI - coreboot, for example, does not reject AI-generated contributions to its repository, and other projects like EDK2 are the same. Even many GNU projects now are starting to accept AI code, or otherwise have lax policies about the use of AI, especially for auditing and code review. I can state, quite reliably, that I am against the existence of LLM technology in its current iteration, but that’s all I can do reliably. If I do, on the other hand, want to have a Libreboot project that is intellectually viable to maintain, then I can’t just go forking every project under the sun. We as small downstream distribution maintainers have this same problem everywhere.
There is resistance. For example, here’s a list of projects that actively try to “de-slop” projects:
Libreboot generally recommends avoiding AI. In a project like ours, which relies extensively on the product of reverse engineering, use of AI is all too tempting as it generally makes headwinds in that world. Any superficial benefits given now by such use may only be temporary; the AI alone cannot know what everything means, and a human being still needs to have a vast base of knowledge in precisely what sort of thing they are reverse engineering. However, it is also true that a lot of reverse engineering work is being done much faster these days because of AI, especially reverse engineering based on the decompilation of compiled code; Ghidra databases can be fed to AI chatbots these days and it can actually get to work on a lot of them, helping the hacker to understand entire blobs much faster than before. We are talking about productivity gains at several orders of magnitude. Projects that might take a year now take months, and months become weeks.
This is true specifically of reverse engineering. It has been my experience that AI otherwise tends to slow the user down, because it reduces the amount of work they actually do while increasing input - you talk to the bot. The bot makes a lot of mistakes, and your very thought process behind how everything is done may weaken over time. It has been found in several studies, such as the one I linked, that AI actually reduces productivity in skilled hackers using it, but this pertains to normal software development tasks. Reverse engineering by hand is tedious, incredibly tedious, so if an AI is only used for the initial decompilation of a binary blob but the human author otherwise rewrites code themselves, then this might actually constitute an acceptable use of AI. If it means that more and more binary blobs are replaced with free code, and the overall rate of technical progress actually increases within that specific context, then I would regard it as in inherently Good Thing. I can say this, despite my other stances - and I still think the technology is bad. This also does not discount the fact that such work may still be performed without it. Even a bad tool can be used well.
I can say credibly that there are several initiatives underway in the coreboot project right now that would not be possible without AI. This does not remove the fact that AI is generally a liability, for reasons already discussed and for reasons that I will discuss, but I ask: should I start rejecting good work if it was done with AI, for the sake of purity? The mission of the Libreboot project is to promote free software to users everywhere, and to make it more accessible for non-technical users, within the realm that is boot firmware. If the reality is that extensive volumes of work are being done now with AI, then where does that lead me? Do I simply ignore it all?
On AI-based reverse engineering though:
The problem again though is that this still requires human knowledge, and thus such benefits are indeed superficial; if young people today are leaning more heavily on use of AI, then there will be fewer opportunities for AI to train on the result of such reverse engineering work in the future, and there will be fewer people even still who actually understand what they’re doing. The AIs do not actually think; they are basically giant probability engines generating the most plausible answer based on a set of well-established statistics. It’s basically picking the most popular answers. It’s useless when it faces a question it has never been asked before, such is the case with a lot of reverse engineering work. A particular embedded controller might work completely differently to everything else, with no information about it anywhere.
The better and faster these AIs get, the worse they become as more people begin to use them and depend on them entirely. You cannot simply tell an AI with a few prompts to gut a binary and start writing source code from it. That’s not how this works.
In other words: the very reason such use of AI vis a vis reverse engineering may increase productivity in the short term is also why it will kill it entirely in the long term. That’s what AI does. It seems powerful at first, but there is a plausible chance that deskilling of lesser engineers will prevent such people from becoming the very skilled engineers to whom it offered such gains in the first place, thereby nullifying all such superficial benefits entirely - and I’m not the only one to make such arguments.
AI did not build the Free Software movement, nor could it. It was trained on our code. LLM bots will regularly perform what are essentially DDoS attacks, traversing websites nonsensically just to scrape a little bit more data. Small time operators face high internet and hardware costs as a result. We all suffer.
AI did not create the free software movement, but it will destroy it. Even these skilled hackers who might use it responsibly today, or with some illusion of responsibility, are going to burn out at some point and younger players will need to take their place. If those young people have been trained with AI, then I believe that this superficial benefit we experience now will just disappear. I reject the concept of AI reverse engineering on a long-term basis for that reason.
I will not reject the use of it, or the use of work based on such effort. I will do what I can to encourage avoiding all use of AI, but I can’t control what people do. If you approach me and tell me that a whole new platform of Intel or AMD hardware is now entirely free software or even partially, and you want to add lots of new hardware to Libreboot, then my only concern is whether or not the work is otherwise solid. I will look at the author, and judge for myself whether the work is any good. I may frown upon the methods used, but I will now repeat:
The mission of Libreboot is to vastly increase the number of people that use free boot firmware. It does this by providing support for as many mainboards as possible, with software such as coreboot. If the result is that more people are able to escape proprietary software, then the end justifies the means - because I’m quite certain that the proprietary software is also full of AI slop.
We can clean up that slop. I will try to avoid it in the first place and convince people to do all the work themselves. It’s also true that I’m probably going to reject nearly any AI-generated work I see, because indeed 98 percent of it is pure garbage - yes.
Why?
Consider that for a moment: we are talking about a technology whose use mostly deskills the users, has been shown to actually decrease overall productivity, despite marketing and a few rare instances such as the very real examples given on this page… consider it. Put it all together and think carefully: is this good? I think not. My argument in this article is that I may like the minority of good things done with it, but I think the technology itself is a liability overall. I know that there are some people who will insist on using it in the shortterm, and perhaps they themselves might have come to similar conclusions.
Consider the counterbalance that is AI-generated bug reports. Many project maintainers are burned out these days just because of the sheer amount of nonsense being sent to them, based on queries that took the contributor a few seconds to write, but the project hours to review and thus wasting their time. The DDoS problem is also very real: my own CGit instance for Libreboot was shut down, because LLM bots kept pounding it and I didn’t see the point in running Anubis. I just shut it down - in other words, my right to choose was restricted. Someone - not even a human - decided that I can’t host CGit on my server.
That’s what a Denial of Service Attack is. Its entire purpose is to make you unable to use your own website or service, forcing you to shut it down. That is happening worldwide as small free software projects struggle to stay online. This is a real problem.
Do not misunderstand me. I may wax lyrical about some of the wonders of AI being used in reverse engineering - and it’s absolutely true that there are works going on in coreboot right now that would not have been possible a few years ago, but the deployment and use of this technology is fundamentally irresponsible. I simply state that I intend to make the best of a bad - and very bad - situation. I know that it is not to last. AI is just bad.
The sad truth is that we can’t successfully or meaningfully fight AI in any way, and hence this policy. There are some other projects out there dishonestly claiming to be against AI, but they will still depend heavily on upstreams and various libraries that contain slop; in nearly all cases, these are overlooked. I can say that I personally hate AI, but will I make Libreboot oppose it outright? If I want to oppose it, then I have to remove it. Opposition is just that: the point of opposition is to replace what is being proposed. If I say that Libreboot will entirely oppose AI, then it has to remove all AI code from every project it uses. If it can’t do that, then it cannot oppose AI. I, its creator, can oppose AI, and I do - and I always will. I will always prefer human-made work. When I see AI slop, I’m probably going to reject it. Probably. I might decide in a few cases that accepting certain slop is better than accepting a binary blob - and then work to replace the slop, just as I would a blob, or find someone to do that for me.
We could fight it, but we need it to happen at an industrial scale. Unfortunately, the Free Software movement has largely accepted use of AI, hence the pickle in which we now find ourselves. The fact that this article needed to exist is a consequence of that.
I can try. I don’t think I will succeed, but I will try. I simply do not have the time to de-slop every single line of code in every single project I use. If you are using any modern Linux or BSD system today, there’s a high chance that you are running AI-generated code on your system. Even if you use one of the more principled operating systems, like OpenBSD, applications from upstreams might (OpenBSD ports tree supports all kinds of software and they can’t patch it all).
And what of these de-slop projects? Noble efforts indeed, but consider this:
They will still only maintain their software for use with Linux systems, despite the fact that Linux has an openly lax policy towards AI these days. Linux kernels at around 7.4 or higher contain copious amounts of AI slop, and it’s only going to get worse. Will they port their software to OpenBSD and leave Linux behind? Hell no. Why would they do that? Linux is most of their userbase in the Free Software world. BSD is nice too, but Linux is where all the traffic is at. Supporting Linux users makes sense.
I think that’s OK, for the same reason I think porting your software to Windows is OK. Giving users a choice of how they run your free code is a good thing. There is no shame in it, but we need to be much more honest about the sheer scale of the problem. We need large projects like Linux to oppose AI outright, to oppose it entirely, and recognise it as the existential threat it is, not only to the Free Software movement but to the health and democracy of human culture worldwide.
I don’t think the anti-slop projects are sustainable, unless they are actual original codebases maintained by the original team and with the agenda already to block AI. If it’s, say, GNU GRUB, forking that is out of the question unless the actual devs device to fork it. There are cases where large projects do succeed in being forked, e.g. Gitea to Forgejo, but it generally almost never succeeds regardless of the context. We can point to efforts like LibreOffice too - another success story, but again, not the norm. We talk about software freedom but the actual practical reality of forking large projects is much more challenging because the infrastructure has to move, projects have to promote themselves and they need the people. One person reverting and maintaining hundreds, thousands or even tens of thousands of patches cannot be done by just a few people - in many such cases, I suspect these forks will fade and fizzle out. That is just a natural consequence in many cases.
This does not necessarily weaken their arguments - and in fact, I would say that I fully support such efforts. I think if you want to fork a major project like say, a large bootloader, a webserver or anything else of that nature, then you must command my absolute respect - and you do, but I will not follow you. I will do what I can, but I look at the overall effect of my actions and decide accordingly. I will choose which problems to tackle and which to ignore, and it’s quite likely that I will ignore nearly all AI-generated patches in upstreams that I package in Libreboot.
I care deeply about this issue, but I will lie neither to myself nor my users. I am one person. I will do what I can and no more. If that ends up being nothing at all, so be it. Even in the most conservative tendency of just reverting a few minor patches here and there, I may find over time that I now maintain thousands upon thousands of tiny, irrelevant patches, and then that becomes a maintenance chore - I end up forking these projects. I won’t try to do what I know will not succeed, so it’s probably that I will ignore nearly all such issues entirely. I think that this is going to be the most common response amongst other maintainers as well, especially those of aggregate software distributions like Libreboot - think also about Linux distros, as the same arguments exist there, at en even higher scale and they have the exact same problem. Some projects handle it better than others.
I think I strike the right balance here, but I could be wrong about all of it.
Libreboot will not promise to remove all AI-generated code - and may, under certain circumstances that I’ll get into next, accept it.
Hence this policy: the AI Slop Reduction Policy. Libreboot initially implemented a total ban on AI as of Libreboot 26.11, but this was short lived. I quickly reconsidered and concluded that such a policy is dishonest. What I can say now is that I will attempt to ban all AI. I will attempt it, but I do not promise to get it all. I can’t stress this enough. It is impossible for me to credibly make such a promise, for a project of this scale.
There are countless examples of where it is still possible to do without AI. Indeed, projects can avoid it entirely if they wish, but Libreboot will not lie to its users. I will not lie.
If you give me a big red button that makes it happen, I will absolutely push that button without a second thought. I remember the world that existed right up until about the year 2020/2021. That was a sane world, and a just world. The one we’re entering now is much stranger and scares me. I worry about AI because I see exactly where it’s going. I’m fortunate enough to have received an education before the AI era. I still have my own mind intact, and I know I won’t use AI myself. I have seen firsthand that my fears are not mere theory; I have seen it deskill people, fast. I’ve seen people who have been programming for over a decade lose all of their skill overnight. If they do break out of the fog, it might take them a few weeks to recover. It’s still relatively easy to do, but I fear that AI will become more entrenched and escaping it will basically become a fantasy in the future.
The brutal truth is this:
Coreboot contains slop these days. That is simply a fact. Some of it isn’t even that bad! That’s the other nuance that a lot of people miss: AI, the technology, and how its implemented and governed, and generally monopolised, is bad, and it’s true that the vast majority of people using AI to write code are writing very bad, often logically wrong or otherwise nonsensical code. That is why we call it AI slop. Because it’s slop. AI is getting better these days, but tens of thousands of lines of code that you write in a week is still slop. It’s regurgitated, kentucky-fried slop. It may even make you feel good. People feel powerful when they can, for the first time, tell a robot what they want and get it immediately. Even if something may not work quite right, it might do a particular job that someone wants.
Do I want to fork coreboot? No. I want to work with coreboot. I want to continue using coreboot, so that Libreboot is feasible for me to maintain. I would need a team of about 100 people, probably more, to properly fork coreboot and remove all of the slop. Upstream doesn’t care. Their own policy only says that the contributor accepts responsibility for code, under the normal sign-off policy. They make no statement against AI whatsoever.
To be clear, because I need to be absolutely 100% crystal clear to the reader: Libreboot contains coreboot. Ergo, Libreboot contains AI-generated code. Coreboot accepts AI-generated patches in its tree, and Libreboot distributes coreboot, which means that the current Libreboot release contains AI-generated code. This is a fact.
How do I fight that? Easy: revert it all. Problem solved. Really? What if a patch isn’t marked as Assisted-by whatsoever? What if I have to remove tens or hundreds of mainboards because an entire platform is AI now? Perhaps things are not that bad now. Indeed, I believe that with a project like coreboot it’s probably even less of a concern in practise, because the vast majority of patches are still going to be human-written, and the slop ones will probably be rewritten. Coreboot does that all the time. It actively encourages everyone to audit its code, and it is quite receptive to rewriting entire sections of the code. For example, it completely rewrote its resource allocation logic several years ago, ditto generalised bootblock logic and a whole host of other things.
Then the elephant in the room: EDK2. Do you think modern EDK2 doesn’t contain AI-generated code?
There are cases however where AI is being used to good effect. For example, coreboot has patches upstream for libre Intel FSP replacement on Kabylake platform; this could be used to replace the Intel FSP blob for raminit and silicon init on a ThinkPad T480. Looking at the code, it’s clear that it was an AI being given a Ghidra database. However its author actually has been doing coreboot work for years, even pre-AI and is highly competent - and I am quite sure that I will be merging his work in a future Libreboot release. That is just a fact. I fear for the sanity of people using this technology, and that they will still become deskilled and lose their intellect over time - the brain can be lost quite easily. Deskilling is a real thing.
I will try. In most cases, I will try. I will. I absolutely do not want AI in any of my projects. I also do not want binary blobs, which led me to another dilemma that I will also solve, now:
In that example, I know for a fact that Arthur Heyman is a highly skilled engineer who understands every aspect of the code he is publishing, on his various projects. He published this article about it:
https://blog.aheymans.xyz/post/thinkpad_x61/
In it, he talks about and defends his use of AI reverse engineering the Lenovo ThinkPad X61 which he then added to coreboot. I basically agree with every point he makes in his article; I myself would not be so quick to promote AI as a positive, but his basic argument seems to be that it exists anyway so why not use it? We might not have gotten ThinkPad X61 support otherwise. He has ported a lot of other boards himself over the years, without AI, and I’ve every confidence in his ability. He was a member of the Libreboot project many years ago and he is credited heavily on the contrib page for that.
I think the world was better off without AI, but I can read his article and regard it as entirely reasonable. I trust that Arthur understands everything he commits and has full control, and that the code will be continually refined over time, as has always been the case in the coreboot project which takes code auditing seriously. This is the same person who recently published sources replacing Intel FSP on 8th gen hardware:
https://review.coreboot.org/q/topic:%22kabylake_native%22
If this were any random person with very little knowledge of how this hardware generally works, then I would regard such work indeed as most likely rubbish and unsuitable for merge. You can’t just push a button and presto, board port complete. I mean. You also literally can do that sometimes, it’s called util/autoport - I digress. The point is that I trust Arthur’s work. I trust the man.
This is non-trivial work. If I were to stick to my guns and never merge AI assisted code, then we would continue to provide binary blob Intel FSP on e.g. ThinkPad T480. I will be entirely honest: I plan to merge this code in a future Libreboot release. It is. I demand it, and I will be the one to merge it. I will do it with joy, and without delusion; I will know, precisely that what I’m merging is assisted by-product of a tool whose engine is derived entirely on the infringement of copyrighted works, most notably GPL-licensed works, and possibly others. Although I didn’t see any written confirmation that AI was used for this work, I know that reverse engineering the FSP as quickly as Arthur did is not possible for a human being, especially writing the tens of thousands of lines of code in all those diffs.
I mean no disrespect, especially to Arthur if he’s reading this. All I mean is that without AI, doing this amount of work for one person this fast is literally impossible, even if you have an IQ of 505. Unless perhaps he has been doing it for years, and I see no indication of that either - this does not diminish the work. I would regard Arthur as its author entirely. He is the one who has to understand it, and modify it if that is the case - he has to own the code. He is responsible for it. I’ve no doubt whatsoever that Arthur is a highly skilled engineer who understands every aspect of the code that he has published.
I have discussed this particular patchset with several other coreboot developers and we all came to the exact same conclusion regarding its likely origin, though I will obviously not mention names. Regardless, I think the work is good and will benefit everyone; I do not want to ship an FSP blob that has been bitrotting since 2017. I plan to merge Arthur’s work, and I will do so quite happily. I will not give it a second thought, and I will evn thank him as I do now.
I can’t prove that it’s AI assisted, unless I see him literally mark it as such. Even if it was, and I believe it was, I plan to merge this code. Because then I comply with Libreboot’s other reduction policy: the binary blob reduction policy.
Not only can I not prove it, I don’t care: it’s going in. I am monitoring this review series intently, and this code will be merged in a future Libreboot release. There is another revolutionary series of works - that I will be tight lipped about for now - from another person I trust, which is very much the same. Yes. You’ll just have to wait.
This is only one small example, and that is the reality I’m afraid. I will not do a disservice to my users by denying them the opportunity to make use of otherwise good work - if I can help someone replace proprietary software on their computer, I will, and the manner in which it was developed is of no concern to me. The four freedoms are absolute.
Another example: Libreboot openly promotes the use of ReBIOS CSM module with EDK2, for legacy BIOS support. When I checked, at around the time of the Libreboot 26.11 release, nearly all (possibly all) of the commits in ReBIOS were AI-assisted. All. Here’s something funny:
https://github.com/ReBIOS-Project/ReBIOS/issues/29
That was a Libreboot user reporting a bug to ReBIOS. You may see there that an LLM triaged and fixed the bug. Magic. I will still recommend use of ReBIOS. Why? Because ReBIOS is available with source code under a free software license, and proprietary CSM is much worse - and likely slopped anyway.
I’ve read the ReBIOS source code myself. It’s absolutely solid code. It’s fine. It is actively maintained and it boots Windows XP. That’s what it’s for. I’ve looked into the people maintaining it, and I’m once again confident that this is a competently maintained codebase. It would probably be even better without AI, but I have no problem recommending its general use. I use it myself.
I’m not convinced about legality yet, as I see no indication that it’s even feasible to hold the sheer number of projects using AI now to account in the courts; indeed, I see that there may be a precedent in the future to regard AI-generated output as essentially public domain. If that happens, it basically means that copyleft is already dead. A lawyer friend of mine showed me this article, arguing that it will probably fall down to something like this:
https://en.wikipedia.org/wiki/Monkey_selfie_copyright_dispute
Clanker monkeys of the world, unite.
I initially imposed an outright ban on AI in the Libreboot 26.11 release, but this was pure folly. I will just do what I can to resist it, without making any promises one way or the other. Free software comes first, no matter who the author is, human or otherwise.
Some short-term good may be had, but the general output is pretty bleak. I basically think this problem will fix itself somehow over the next thirty years. I don’t think AI is going away. Our job as maintainers is to be as honest as possible about what we can and cannot do, and it looks like AI is basically taking over the world. My conclusion in all of this is that we need to think about re-claiming AI for ourselves; if training were done using Free Software, in a distributed peer-to-peer fashion and in a manner entirely transparent to the public, like seti@home for example, it would also remove the need for centralised datacentres and we could more easily reduce the baseload demand as use of renewable technologies like solar and wind become far more practical. I definitely want to see an end to the proprietary chatbots that currently exist - local AI models are already a thing, but they’re still in their infancy and the actual training is completely out of our control for the time being.
Really though, I just hope we one day return to a world where human beings think for themselves. It remains to be seen what the full long-term consequences of AI are, but I basically think nothing good will ever come of it. Any short term benefits we might perceive now will mean nothing in thirty years. I’m not optimistic at all.
I basically think we’re completely screwed.
Markdown file for this page: https://libreboot.org/news/clankers.md
This HTML page was generated by the Libreboot Static Site Generator.