- Updated: March 27, 2026
- 48 min read
AI‑Generated ext4 Driver Sparks Legal Debate in OpenBSD Community
LWN.net News from the source ContentWeekly EditionArchivesSearchKernelSecurityEvents calendarUnread commentsLWN FAQWrite for us User: Password: | | Log in / Subscribe / Register Vibe-coded ext4 for OpenBSD [LWN subscriber-only content] Welcome to LWN.net The following subscription-only content has been made available to you by an LWN subscriber. Thousands of subscribers depend on LWN for the best news from the Linux and free software communities.If you enjoy this article, please consider subscribing to LWN. Thank you for visiting LWN.net! By Jonathan CorbetMarch 26, 2026 A number of projects have been struggling with the question of which submissions created by large language models (LLMs), if any, should be accepted into their code base.This discussion has been further muddied by efforts to use LLM-driven reimplemention as a way to remove copyleft restrictions from a body of existing code, as recently happened with the Python chardet module. In this context, an attempt to introduce an LLM-generated implementation of the Linux ext4 filesystem into OpenBSD was always going to create some fireworks, but that project has its own, clearly defined reasons for looking askance at such submissions.It all started on March 17, when Thomas de Grivel posted an ext4 implementation to the openbsd-tech mailing list. This implementation, he said, provides full read and write access and passes the e2fsck filesystem checker; it does not support journaling, however. The code includes a number of copyright assertions, but says nothing about how it was written. In a blog post, though, de Grivel was more forthcoming about the code’s provenance: No Linux source files were ever read to build this driver. It’s pure AI (ChatGPT and Claude-code) and careful code reviews and error checking and building kernel and rebooting/testing from my part. There were a number of predictable concerns raised about this code, many having to do with the possibility that it could be considered to be a derived product of the (GPL-licensed) Linux implementation. The fact that the LLM in question was almost certainly trained on the Linux ext4 code and documentation does not help.Bringing GPL-licensed code into OpenBSD is, to put it lightly, not appreciated; Christian Schulte was concerned about license contamination: I searched for documentation about that ext4 filesystem in question. I found some GPL licensed wiki pages. The majority of available documentation either directly or indirectly points at GPL licensed code. In my understanding of the issue discussed in this thread this already introduces licensing issues.Even if you would write an ext4 filesystem driver from scratch for base, you would almost always need to incorporate knowledge carrying an illiberal license. Theo de Raadt, however, pointed out that reimplementation of structures and algorithms is allowed by copyright law; that is how interoperability happens. One should not conclude that De Raadt was in favor of merging this contribution, though.From the OpenBSD point of view, the copyright status of LLM-generated code is indeed problematic, for the simple reason that nobody knows what that status is, or even if a copyright can exist on that code at all. Without copyright, it is not possible to grant the project the rights it needs to redistribute the code.As De Raadt explained: At present, the software community and the legal community are unwilling to accept that the product of a (commercial, hah) AI system produces is Copyrightable by the person who merely directed the AI. And the AI, or AI companies, are not recognized as being able to do this under Copyright treaties or laws, either. Even before we get to the point that the AI’s are corpus-blenders and Copyright-blenders.So as of today, the Copyright system does not have a way for the output of a non-human produced set of files to contain the grant of permissions which the OpenBSD project needs to perform combination and redistribution. Damien Miller said something similar: Who is the copyright holder in this case? It clearly draws heavily from an existing work, and it’s clear the human offering the patch didn’t do it. It’s not the AI, because only persons can own copyright.Is it the set of people whose work was represented in the training corpus? Was the it the set of people who wrote ext4 and whose work was in the training corpus? The company who own the AI who wrote the code? Someone else? We don’t know. The law hasn’t caught up to the technology yet and we can’t take the risk that, when it does, it will go in a way that makes use of AI-written code now expose us to legal risk.These words did not resonate entirely well with de Grivel, who refused to retract his copyright claims on the machine-generated code. He also is clearly pleased with the kinds of things one can do with LLMs: We can freely steal each other in a new original way without copyright infringment its totally crazy the amount of code you can steal in just 1h. What took 20 years to Bell labs can now be done in 20 hours straight.The conversation went on for some time, but the result was never really in doubt; De Raadt made it clear when he said: “the chances of us accepting such new code with such a suspicious Copyright situation is zero”. In the above-mentioned blog post, de Grivel added a note on March 23 that he would respond by removing all of the LLM-generated code, leaving only code that he has written himself.After this episode, though, convincing others that he really did write any subsequent versions on his own may be an uphill battle. He acknowledged that “forking OpenBSD” might be easier. The number of people who have concluded that they can have an LLM crank out thousands of lines of code and submit the result to the project of their choice is growing quickly.Needless to say, these people are not always diligent about documenting the provenance of the work they are submitting in their own names. There may well come a time when it turns out that even the sharp eyes of OpenBSD reviewers are unable to keep all of it out of their repositories. All of this code is setting some worrisome potential traps for the future. As Tyler Anderson pointed out, the price of these tools is unlikely to go down as development projects become more dependent on them.Who will maintain this code, when its original “author” does not understand it and has no personal investment in it, is unclear at best. And if there is, in fact, a potential copyright problem inherent in this code, there will have to be a lot of scrambling (or worse) when it comes to light. Given all of that, it is unsurprising that many projects, especially those with longer time horizons, are proving reluctant to accept machine-generated submissions.Index entries for this article KernelFilesystems/ext4 to post comments I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 15:23 UTC (Thu) by koverstreet (✭ supporter ✭, #4296) [Link] (27 responses) The whole point of free software, way back, was so we could stop fretting over who has the rights to what and just read and work on each other’s code.This obsession with copyrights between different free software ecosystems – who put the lawyers in charge? Is anyone asking the obvious question? Is the code any good? Skimming through it, it looks decent – if you like C written in a style 25 years ago; it is very reminiscent of ext* code. The data structures are not terribly clear, there’s no real separation of concerns between higher level algorithms and the low level data structures. We can do far better today.The bigger concern with LLMs writing code is maintainability and refactoring. Claude has a massive blind spot when it comes to refactoring and cleaning things up as you go; Anthropic has tuned it very heavily for “just get something working autonomously (without any concern for what comes after)”, which means that in practice the first version will look pretty good – but then if you’re too hands off with continued development it will rapidly go off the rails.It understands on a conceptual level structural duplication and how to fix any sort of code smell you might point out to it, but you have to keep a tight hold on the reins or it’ll insist “nah, this is clean code, everything is fiiiiiiine”. Amusingly, Qwen3.5-27b is far better at spotting structural issues and seems to have better taste in design. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 26, 2026 16:40 UTC (Thu) by bertschingert (subscriber, #160729) [Link] (2 responses) > The bigger concern with LLMs writing code is maintainability and refactoring. Not to mention the more fundamental question of whether writing new code is the right way to solve a problem.This snippet from the conversation seems pretty important: On Mon, Mar 23, 2026 at 08:32:43AM +0000, Stuart Henderson wrote: > On 2026/03/23 03:45, Thomas de Grivel wrote: > > I did check everything and it did look like ext2fs code with minor > > modifications that I fully understand to support ext4 extents and > > checksums it’s not that big of a deal : ext4 is just ext2 with 64 bit, > > different checksums and extents tree walking. > > Why standalone rather than done as diffs to ext2fs?If a tool makes it extremely easy to generate lots of code, then people will tend to solve problems by generating lots of code. I already see this with my coworkers writing multi-thousand-line programs to do something that could be done with a short bash script combining existing tools. As a sysadmin, every new program I have to read, understand, and maintain is another burden. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 26, 2026 21:45 UTC (Thu) by ballombe (subscriber, #9523) [Link] (1 responses) > If a tool makes it extremely easy to generate lots of code, then people will tend to solve problems by generating lots of code. I already see this with my coworkers writing multi-thousand-line programs to do something that could be done with a short bash script combining existing tools. As a sysadmin, every new program I have to read, understand, and maintain is another burden. Exactly.I start to receive LLM-generated patches that are the equivalent of the coreutils maintainers receiving a patch labelled “cp: add –url option to copy remote files, like cp –url https://lwn.net/ lwn.html”, the end result being that every software will blow up to include a unmaintainable and barely working reimplementation of every other, with both an editor AND a MTA. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 26, 2026 22:43 UTC (Thu) by Wol (subscriber, #4433) [Link] Haven’t you forgotten a complete functional copy of emacsOS? 🙂 Cheers, Wol I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 17:03 UTC (Thu) by Heretic_Blacksheep (subscriber, #169992) [Link] (11 responses) Who put the lawyers in charge? Everyone in a given society. That is, the majority of any given democratic or republican system of governance.That’s not a useful debate to have. When you live in a society, you have to live by that society’s rules or face the consequences even if you disagree with the rules. As a traditionally persecuted minority in the country I live in, I’m quite familiar with the risks of opposition to The Rules. The OpenBSD stance towards copyright is the only legally sane one currently. The output of LLMs is currently ambiguous at best in any given country.The output of pure LLM driven churn cannot be held by the LLM. Who actually does is as ambiguous as always with any mixed parentage work has been. The USL settlement is informative but not entirely on point. The monkey selfie lawsuit is also informative, but not fully on point either. Copyright law pivots on the *exact* details of any given case. That means any given situation depends on points of law specific to the very specific circumstances of the case. Even if the results of Bartz, et al.v. Anthropic and Kadrey v. Meta say the situation on training data is mixed but not entirely on point for all end use. It depends on the legality of how the training data was originally obtained. But so far, no one has ventured to sue an *end user* for the output of an LLM to find out what liabilities the action of using an LLM to generate content – especially when someone openly admits to using an LLM as an end run around copyright, as in this case.That’s immediately willful infringement should any ruling (or future law) go against the entities trying to get away with it. OpenBSD is right to be conservatively wary. The point of free software wasn’t to do away with copyright. It was effectively one of subversiveness, but free software actually *requires* copyright to make any attributive or forced contributory license function.If the goal of the ideal is put into play, then the actual goal of free software was never to subvert software licensing, it was to democratize the building of software and the free reign of the hacker ethic. That is the ethic of free inquiry into methods, practices, and procedures implied by the ownership of (and some would say the existence of) an object or group of objects. In short, it’s to utilize the right to tinker on things one owns.That inquiry is not necessarily in full tension with copyright, and it’s not opposed to the use of LLMs as a democratization vector. It’s just that the current legal situation with the output of LLMs is unsettled so it’s right to be wary of commercial LLM services where the output is en mass directly incorporated into existing or the whole part of a creative project.It’s even more prudent to be wary of the output of anyone that expresses that their entire goal for the existence of a project is to break their understanding of the law! In effect, the sole purpose is to facilitate the piracy of software rather than to further the hacker ethic! I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 26, 2026 17:24 UTC (Thu) by Wol (subscriber, #4433) [Link] (1 responses) > The USL settlement is informative but not entirely on point. Hmm . USL was basically “the emperor has no clothes, but there’s no point rocking the boat”. Unix was written when (under US law) software was protected by trade secret, not copyright.AT&T removed all copyright notices, including debatably legal BSD copyrights, and very definitely legal UK and Aussie copyrights (the copyright status of software has never been questioned there, afaik). Then when they sued the Regents for copyright violations, they slapped their own copyrights on everything including stuff they’d stripped other peoples’ copyrights off of.Once the Regents convinced the Judge that AT&T copyrights had been slapped on the Regents’ code, and worse AT&T had no idea what was their own code and what was other peoples’, the wheels fell off the lawsuit, and for the sake of face, the Regents and AT&T agreed never to discuss the matter. What USL, and Novell, and Caldera got when they bought the assets was a quitclaim – “whatever we have, you’ve now bought. Good luck proving what it was”.So I don’t actually think that lawsuit says anything about LLM copyright (and in case you’re wondering, the “secret” AT&T/Regents settlement is now in the Public Domain – it was posted to Groklaw.) Cheers, Wol I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 18:38 UTC (Thu) by Heretic_Blacksheep (subscriber, #169992) [Link] I said it’s informative, not entirely on point, which is effectively what you’re expanding upon.In legal parlance that means some of its legal principle is applicable, not all of it. “On all points” means the legal principles have direct bearing on a case in substantially all details. “Not entirely on point” means there’s a point or two that are influential, but has no direct bearing as in an all points precedence.The applicable parts are those files that could hold copyright in an otherwise mixed and very muddled environment of pre-Copyright Act files (pre Copyright Act of 1976*) and those with muddled authorship that couldn’t be clearly established but substantially or wholey written after 1976. The few files that were clearly copyright AT&T were removed and replaced in the BSD with clean replacements after the 1994 settlement.To imply that all of Unix was under trade secret rather than copyright law is not true since Unix (both AT&T derived and Berkeley derived) clearly evolved after 1976. For example, Unix System V was released in 1983 and would very clearly have been largely copyrightable if provenance had remained undiluted. The details of a case matter. Some matter more than others. *Apple v. Franklin (1983) ruled machine code, and by extension human readable code, are in fact covered under the Copyright Act.Google v. Oracle (2021) affirmed header files are also covered by the Copyright Act. That means all source code, functional and descriptive, since 1976 is potentially covered by federal copyright. Not that there weren’t copyright laws before 1976, the states managed that with contradictory legal regimes that may or may not have covered human readable source code. It was the complexity, contradictions, and inconsistency of state copyright regimes that prompted the 1976 Act.I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 20:00 UTC (Thu) by koverstreet (✭ supporter ✭, #4296) [Link] (3 responses) > Who put the lawyers in charge? Everyone in a given society. That is, the majority of any given democratic or republican system of governance. That’s not a useful debate to have.When you live in a society, you have to live by that society’s rules or face the consequences even if you disagree with the rules. As a traditionally persecuted minority in the country I live in, I’m quite familiar with the risks of opposition to The Rules. I can’t get behind this “we all agreed to the rules and we all must follow them” line of thinking, with not a thought given to how they’re serving the free software worlds and the commons.I want a healthy free software ecosystem, and I don’t see how this helps. And I think you’re taking a narrow and pessimistic view of the legalities of LLMs, and I honestly can’t make any sense of it. If there are going to be legal issues with using LLMs trained behind closed doors, that’s an issue for the entire software world, not just open source – and honestly if that ever came to pass it’d be a far bigger issue for the corporate world.Corporations might sue each other for infringement, but open source projects? But given how the eagerly the rest of the world is charging (headfirst, with nary a care in the world) into LLM driven development, somehow I do not think that’s the issue that’s going to bite the open source world. I think that particular ship has sailed. The only issues on my mind are – how do we teach AIs to write good maintainable code? How are we going to collaborate?I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 22:21 UTC (Thu) by Heretic_Blacksheep (subscriber, #169992) [Link] (2 responses) I didn’t say “We all must follow them” I said “When you live in a society, you have to live by that society’s rules or face the consequences even if you disagree with the rules.” There’s a gulf of difference.One can have a debate on the justice of those rules and laws, but ask anyone that has openly defied those laws and been arrested, marginalized, or otherwise attacked as a matter of protest whether there are consequences to doing so. However, I don’t believe the wider issue is a topic for LWN. I put the comment there because the OP questioned who put the lawyers in charge. The answer is obvious. The society did, for good or ill. That’s just the world we live in.I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 22:37 UTC (Thu) by mb (subscriber, #50428) [Link] (1 responses) > The society did Wrong. Individuals do this. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 22:40 UTC (Thu) by koverstreet (✭ supporter ✭, #4296) [Link] yeah, and I think “the society has decided!” has become an excuse to shut off critical thinking. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 20:14 UTC (Thu) by ceplm (subscriber, #41334) [Link] (4 responses) > Who put the lawyers in charge? Do you understand that this is the question which has been with humankind since at least since eighteen century BCE (Code of Hammurabi) and probably much earlier, right? Do you think that LWN is the best place to solve it? I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 22:08 UTC (Thu) by koverstreet (✭ supporter ✭, #4296) [Link] (3 responses) If you’re jumping all the way to the Code of Hammurabi, I think you might be proving my point for me 🙂 Not everything has to be, or should be, a legal issue. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 26, 2026 22:31 UTC (Thu) by Heretic_Blacksheep (subscriber, #169992) [Link] He’s not. He’s pointing out that the laws of a society exist because that society believes they need them to order their social interactions. The earliest evidence of such a full legal framework is the Code of Hammurabi. He’s not making your point. Copyright is a legal question because modern societies believe creative works deserve some kind of legal protection.You need not agree, but you can’t ignore the question when working with someone else’s implementation of ideas without potential consequence. Understandably OpenBSD wishes to avoid those legal consequences if at all possible. Wishes for utopias without lawyers and legal code are wishful thinking and irrelevant to the present. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 27, 2026 1:19 UTC (Fri) by Kluge (subscriber, #2881) [Link] (1 responses) > Not everything has to be, or should be, a legal issue. Ownership of anything is inherently a legal issue. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 1:24 UTC (Fri) by koverstreet (✭ supporter ✭, #4296) [Link] And here I was thinking open source and free software were about user freedom and just being able to tinker on things.Silly me 🙂 I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 18:44 UTC (Thu) by sionescu (subscriber, #59410) [Link] (8 responses) > This obsession with copyrights between different free software ecosystems – who put the lawyers in charge? The constitution. We live under the rule of law. > The bigger concern with LLMs writing code is maintainability and refactoring. No.The bigger is that it hasn’t still been legally established who owns their output. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 21:24 UTC (Thu) by Wol (subscriber, #4433) [Link] (7 responses) > No. The bigger is that it hasn’t still been legally established who owns their output. It still hasn’t been legally established who owns the output of a Natural Intelligence either!Just treat the output of Artificial Intelligence the same way you treat a Natural Intelligence. Apply some of that mythical ingredient called Common Sense 🙂 Cheers, Wol I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 21:28 UTC (Thu) by sionescu (subscriber, #59410) [Link] (6 responses) Don’t be silly. Copyright and contract law establishes that pretty decidedly.I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 26, 2026 22:42 UTC (Thu) by Wol (subscriber, #4433) [Link] (5 responses) So is my output copyright me? That is NOT clear. What if I do a Mozart, go to a concert, and write out the score from memory? What if I do what a lot of musicians do, and extemporise a piece where I have no idea where the inspiration came from? If anywhere? There’s been a few lawsuits over that!Cheers, Wol I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 10:37 UTC (Fri) by anselm (subscriber, #2796) [Link] (3 responses) What if I do what a lot of musicians do, and extemporise a piece where I have no idea where the inspiration came from? If anywhere? There’s been a few lawsuits over that!The general stipulation, according to the Berne Convention, which most civilised countries subscribe to, is that new “works” which you – as a human being, capable of creative expression – produce are yours from the moment you “expressed” them. If other people feel that your work seems to be ripped off from their work, they are free to litigate this after the fact. As a musician you probably know lots of cool tunes by other people and are polite enough not to try to pass those off as your own.OTOH, it is unlikely for you to know every currently-copyrighted tune in the repertoire in order to be able to avoid those in your own work, which is why there are lawsuits about this sort of thing. Although in fairness it must be said that some of these lawsuits are based on fairly fleeting resemblances of someone’s very obscure prior work to someone else’s more recent very successful smash hit, and this sometimes carries a whiff of people trying to capitalise on others’ success.I presume that adjudicating such cases can be quite tricky. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 11:39 UTC (Fri) by Wol (subscriber, #4433) [Link] (2 responses) So we just have to apply the exact same – FUZZY – rules to AI output. Which makes me think this vibe-coded driver is actually a derivative of the GPL’d Linux driver .Cheers, Wol I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 13:19 UTC (Fri) by anselm (subscriber, #2796) [Link] So we just have to apply the exact same – FUZZY – rules to AI output. For the time being, anyway, AI output flunks the “creative expression by a human being” prerequisite for being able to be copyrighted at all – see the monkey-selfie case.So whether a copyright on the AI output belongs to the AI, its human prompter, the company which pays the human prompter’s salary, or the company which put the AI together in the first place is a moot point because there is no copyright to be allotted to begin with.This can only change if the legal fundamentals of copyright are amended to declare that AI is capable of “creative expression” in the same way human beings are (unlikely, but what do I know), or else if you convince the court that you have put so much of your own creative expression into the work that it should qualify for copyright on those grounds even though an LLM was also involved in its creation in some ancillary capacity.IOW, the “viber” the code, the less likely it is to be eligible for copyright. Having said that, if you publish code that your AI has produced (with more or less additional input by you), you still run a risk that that code is substantially similar to some other existing work up to a point where the author of said work could reasonably conclude that they have been ripped off, and a lawsuit ensues.(Obviously, the same thing could happen with code that you have personally written completely from scratch.) With strictly human-authored code, the usual defense is to argue that you (the author) have never actually even seen said work and that any resemblance must therefore be coincidental (possibly you worked off an abstract specification written by somebody who has seen the other work, i.e., “clean-room engineering”, but that is generally considered acceptable).Whether that defense will in fact fly in your particular case is anyone’s guess, but the same argument is way harder to make if part of your work has been generated by an LLM that was potentially trained on all the visible source code on the Internet but nobody knows for sure what actually went in there. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark.Posted Mar 27, 2026 13:20 UTC (Fri) by daroc (editor, #160859) [Link] It is, in a sense, not really relevant whether a judge would or would not determine that the new driver is a derivative of the Linux driver; the point is that the OpenBSD project has no particular wish to be dragged into a complicated, fuzzy, expensive legal battle over this code. And that remains a reasonable desire regardless of the actual legal status of the work.I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 20:06 UTC (Fri) by wahern (subscriber, #37304) [Link] > What if I do a Mozart, go to a concert, and write out the score from memory? As an aside, that reminds me of an interesting bit of history in the famous Sony Betamax US Supreme Court case.In Justice Stevens first draft opinion, he wrote, “It would plainly be unconstitutional to prohibit a person from singing a copyrighted song in the shower or jotting down a copyrighted poem he hears on the radio”. That second scenario, transcribing a poem, gets black letter copyright law dead wrong, at least in the sense he meant–his original opinion argued the Copyright Act didn’t apply at all to such private, personal use.IIRC, the other justices had to correct him that that is the very definition of copyright infringement, and the definition of infringement per se never distinguished private from commercial. Fair Use would protect a person from liability (and the First Amendment would likely demand it), but it’s nonetheless infringement, per se; Fair Use is a *defense* to infringement. > What if I do what a lot of musicians do, and extemporise a piece where I have no idea where the inspiration came from?In the US, at least, you generally prove infringement by showing 1) substantial similarity (to copyrighted material) and 2) access to the copyrighted work. It doesn’t matter your intentions or whether you even knowingly copied or made a derivative. Access can be proven by showing that, e.g., a work is effectively universally encountered, e.g. Beatles’ songs. As regards AI or anything else, substantial similarity and access are really all you need to show infringement.Showing access in the case of AI is basically a slam dunk for any published work, given large models have been trained on effectively everything that has been digitized. Copyright cases usually turn on the degree of similarity, or other issues, like whether a work was actually protected by copyright, whether copied material was copyrightable, establishing the rights holder, application of Fair Use, etc.But the question whether infringement could occur is the simplest aspect to copyright law, at least in countries where copyright law derives from common law copyright. Your hypotheticals don’t actually pose any sort of challenge to the basic logic of infringement in this regard. The keystone to copyright law is the concept of originality, and answering the question of originality is fundamentally just 1) substantial similarity and 2) access.I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 13:09 UTC (Fri) by Conan_Kudo (subscriber, #103240) [Link] (1 responses) The whole point of free software, way back, was so we could stop fretting over who has the rights to what and just read and work on each other’s code. No, it wasn’t. Nobody considered that at all.And you should know better, it’s literally written on the GNU website what the principles of free software actually are. At the core of it, it was about protecting users from developers concentrating their power and locking things down. This obsession with copyrights between different free software ecosystems – who put the lawyers in charge? Society did. In order to have rule of law, you need arbiters of law. Justices, judges, and attorneys are the backbone of this model.And copyright matters because this is the legal tool to convey intent for the work. And this has been fundamentally recognized in every society on earth. I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 14:41 UTC (Fri) by Wol (subscriber, #4433) [Link] > Society did. In order to have rule of law, you need arbiters of law. Justices, judges, and attorneys are the backbone of this model.And even if your society consisted of one nuclear family ONLY, you still need people to fill those roles (yes, in a society that small, it’s probably just Dad :-). So the only way to escape, is to go off to your own isolated island and live there on your own. This is what I call “emergent complexity”, it just happens of its own accord. It’s a natural consequence of the fact that humans are social animals.Cheers, Wol I can’t help but feeling like if we’re just going to be talking about copyright, we’ve jumped the shark. Posted Mar 27, 2026 15:36 UTC (Fri) by jadedctrl (subscriber, #178426) [Link] > This obsession with copyrights between different free software ecosystems – who put the lawyers in charge? All you need to do is look at the history of copyright disputes and suits that hampered BSD for literal years to see where this “obsession” came from.There are very, very good reasons to worry about being above-board, copyright-wise. This lesson was learnt the hard way. This is a full mgnitude smaller, but still. Posted Mar 26, 2026 17:01 UTC (Thu) by ccchips (subscriber, #3222) [Link] (2 responses) When I first saw ChatGPT and DeepSeek, I got the idea to write an analog clock in tkinter by “vibe coding.” I told the bot what I wanted, how I wanted it to work, and how I wanted it to look.The outcome worked, but I decided just to report the problems and have it revise its own code. When I got it to work pretty well, I decided to ask the bot for revisions. All hell broke loose. Maybe managers would love “vibe coding”, but I don’t. Maybe just to pck up a few pointers on how to do something, without having to do a lot of research, but afte that, there’s still nothng like hard work. I have heard the term “AI garbage,” and I agree with that.Some of the code I got was pretty lame, and frankly, difficult to fix by hand. This is a full mgnitude smaller, but still. Posted Mar 26, 2026 17:04 UTC (Thu) by ccchips (subscriber, #3222) [Link] (1 responses) .and ext4 seems a whole lot more important to get right than a little tkinter project. This is a full mgnitude smaller, but still. Posted Mar 27, 2026 1:21 UTC (Fri) by Kluge (subscriber, #2881) [Link] > .and ext4 seems a whole lot more important to get right than a little tkinter project. If I were to rank projects to be vibe-coded, a filesystem implementation would be dead last. How can one post a kernel driver without understanding it? Posted Mar 26, 2026 17:45 UTC (Thu) by berto (subscriber, #58604) [Link] (1 responses) > Who will maintain this code, when its original “author” does not understand it and has no personal investment in it, is unclear at best.It seems very clear to me that this has zero chances of being accepted. I wonder what the intentions of the submitter were (I cannot read the blog post, the host is down), was he seriously thinking that the OpenBSD community would accept a kernel patch like that? How can one post a kernel driver without understanding it? Posted Mar 27, 2026 18:05 UTC (Fri) by Hibchibbler (guest, #167534) [Link] You need people to push the boundaries in order to see how far the boundaries go.Someone was going to do this one way or another. Economics of commercial AI (LLM) Posted Mar 26, 2026 18:32 UTC (Thu) by gfernandes (subscriber, #119910) [Link] It’s unlikely that the cost of _these particular_ AI systems continues to _remain_ low, is the more likely end game. These AIs are heavily subsidised, taking losses on every user that burns many tokens (quantified by approaching their subscription limits).Add to this fact, the additional fact that they lie about their costs (neither OpenAI nor Anthropic include training or compute costs in their operating costs), as well as their revenues (both present “annualised revenue” on the basis of a good month, projected over a year), you have a situation that is essentially a scam. Both can’t pay the bills they’ve committed to, but are continuing to “sell” services at a discount that ensures they cannot cover their costs. Ever.They seem to remain operational on the almost Crypto-Bros sentiment of “number go up”, where “number” is the amount investors seem willing to commit. But in recent years, their investors (Private Equity) have increasingly got tied up in “loans” from traditional Banks and funds (including pension and insurance). This section will not tolerate fictional income for very long – ultimately affects their interest repayments.So we’re simply looking at a bubble at very close (how close is anybody’s guess) to bursting point. At least if we’re talking about these two and anyone hanging off of their coat tails. As a case in point, the money Disney promised OpenAI, has just fallen off the table with OpenAI pulling the plug on Sora.The Law is already Established; Work for Hire Posted Mar 26, 2026 18:58 UTC (Thu) by roskegg (subscriber, #105) [Link] (16 responses) If you are paying for a subscription to the AI, you retain copyright of the generated work. It falls under “Work for Hire” laws. There is a lot of BSD licensed code up on github right now that was generated by LLMs. For music AI like Suno, this is made explicit. Many book authors are using Claude, ChatGPT, etc, to write books.They are copyrighting them and selling them on Amazon. This is current practice. The Law is already Established; Work for Hire Posted Mar 26, 2026 19:25 UTC (Thu) by roskegg (subscriber, #105) [Link] (1 responses) What Claude has to say on the matter: (and yes, I edited down the response, the original was much more verbose) All four major providers have converged on essentially the same contractual position: they assign output ownership to the user.OpenAI (ChatGPT): As between the user and OpenAI, and to the extent permitted by applicable law, the user retains ownership rights in input and owns the output, with OpenAI assigning all right, title, and interest in output to the user. Anthropic (Claude): Anthropic’s commercial terms enable customers to retain ownership rights over any outputs they generate, and Anthropic will defend customers from copyright infringement claims related to their authorized use of the services or outputs.Anthropic The consumer terms historically were more restrictive — the free version gave users a license for internal, non-commercial use, while paying commercial/API customers get explicit ownership of output. xAI (Grok): As between the user and xAI, and to the extent permitted by applicable law, the user retains ownership rights to user content (both input and output).xAI However, there’s a catch — unlike the other providers, xAI takes an irrevocable, perpetual, transferable, sublicensable, royalty-free worldwide license to use, copy, modify, distribute, and create derivative works from that content. Google (Gemini): Google states it won’t claim ownership over generated content, but acknowledges it may generate the same or similar content for others and reserves the right to do so.Google AI Google also grants itself a worldwide, royalty-free license to use, host, store, reproduce, modify, create derivative works from, and publicly display user content. The Copyright Office distinguishes between minimal human input — where a user enters a simple text prompt and accepts the result — and substantial human creativity, where a creator selects, edits, and arranges AI-generated elements in a way that reflects creative judgment, which may qualify for copyright protection.For code specifically, when a human programmer provides sufficient creative input — such as iterative prompting, editing, and refining of output — copyright ownership may attach to the human author or, through employment agreements, to their employer.The practical implication: the more you treat the AI as a tool rather than the author — debugging, refactoring, integrating, making architectural decisions, writing the logic and using the AI to fill in implementation details — the stronger your copyright claim becomes. The code that survives your review and modification process has a plausible human authorship argument. The code you accept verbatim from a prompt has essentially none.The Law is already Established; Work for Hire Posted Mar 27, 2026 1:18 UTC (Fri) by Kluge (subscriber, #2881) [Link] >What Claude has to say on the matter: (and yes, I edited down the response, the original was much more verbose) I don’t think anyone cares what Claude has to say. I certainly don’t.The Law is already Established; Work for Hire Posted Mar 26, 2026 20:03 UTC (Thu) by rgmoore (✭ supporter ✭, #75) [Link] (3 responses) How people are acting and what the law says are not necessarily closely connected. In this case, the law surrounding copyright an AI is far from settled, and it won’t be settled until the courts have had their say. That may not affect people who are only interested in the short term.People slapping their copyright on AI slop books are just trying to make a quick buck. They won’t care if in a few years the courts rule that those books were never validly copyrighted in the first place, because they will have made their sales and won’t have to refund their customers’ money. OpenBSD, and other people who want to be cautious, are going to wait for the court cases to sort out before they ever accept AI-based contributions.The Law is already Established; Work for Hire Posted Mar 27, 2026 8:58 UTC (Fri) by kleptog (subscriber, #1183) [Link] (2 responses) > In this case, the law surrounding copyright an AI is far from settled, and it won’t be settled until the courts have had their say. Copyright law is reasonably settled. Because copyright law doesn’t care about the process.I think we’re revealing our techie bias when we spend time discussing whether mechanisms: did the AI see copyrighted data, did the user look at the result or not, is the model copyrighted, etc. But none of this is relevant. Whatever code was submitted here, it’s copyright status is determined by whether the code resembles existing code, and not whether it was by AI or by hand. Now, in a court case, the mechanism might be considered evidence but it will never be the factor that decides the case.As an aside, the phrase “until the courts have their say” bugs me. I know there are legal systems that consider the judgements of unelected judges Very Important but I think it should work the other way round. We collectively (democratically) need to decide what is the best way to “promote the arts” and how copyright fits in that picture. And then the courts use those guidelines to resolve specific cases. But to each their own.It’s not yet clear where LLMs are going to be the biggest benefit, but I don’t think a blanket banning is good for anyone. The Law is already Established; Work for Hire Posted Mar 27, 2026 13:10 UTC (Fri) by gfernandes (subscriber, #119910) [Link] You’re right — Copyright law is fairly settled. Consequently, the rather obvious damage which can easily seen by asking questions on problem solutions already on Stackoverflow.com, is almost verbatim reproduction of code. Stackoverflow.com is probably not going to sue — the code is disembodied fragments. But the implications are not. The implication is that any code “learned” by the LLM service provider, can be leaked into a direct competitor unintentionally, and equally verbatim. I’m sure there are at least some copyright implications here. But there are also large implications around stealing business secrets and violating patents.The Law is already Established; Work for Hire Posted Mar 27, 2026 16:51 UTC (Fri) by rgmoore (✭ supporter ✭, #75) [Link] Copyright law is settled in its broad strokes but not in all the details, and it does care at least somewhat about process. In particular, only works of human expression can be copyrighted. By law, works created by computers and non-human animals are considered to lack sufficient expressive content to be copyrightable.That is a critical point when considering the copyright status of any code where LLM played a substantial role in the actual coding. One of the details that still needs to be worked out is how much human contribution is necessary to make something a copyrightable work of human expression rather than uncopyrightable machine output. For questions of copyright violation, the law also considers access to the work allegedly being copied as a factor.It’s easier to prove infringement if the defendant had access to the work they’re accused of infringing. We don’t know for sure- I don’t think the case law is settled- but using an LLM that was trained on the allegedly infringed work will probably count as having access. The typical user lacks sufficient knowledge to testify about their LLM assistant’s training repertoire, so they’d be precluded from claiming they never had access to the work they’re accused of infringing.Even if someone from the LLM company can testify, they will probably wind up admitting that just about any piece of Free Software was included. That alone is reason for people worried about copyright infringement to avoid LLMs. The Law is already Established; Work for Hire Posted Mar 26, 2026 22:20 UTC (Thu) by dilinger (subscriber, #2867) [Link] (9 responses) This assumes that the AI (or the company behind it) has the right to assign copyright for that generated work.If the model has been trained on works that it doesn’t have permission for, does it have that right? The Law is already Established; Work for Hire Posted Mar 26, 2026 22:31 UTC (Thu) by bluca (subscriber, #118303) [Link] (8 responses) The only requirement is that the data used for training was public (ie: didn’t hack a firewall or torrented it, which might seem obvious and yet some companies were caught doing just that) and the publisher did not explicitly opt out.If these requirements are met, then anyone has permission to use it for training. The Law is already Established; Work for Hire Posted Mar 26, 2026 22:53 UTC (Thu) by Wol (subscriber, #4433) [Link] (7 responses) > then anyone has permission to use it for training. But the user has NO automatic right to claim copyright in the OUTPUT! An AI is legally (in Europe) no different from a person – it can ingest copyright works for training purposes.The assumption (not explicitly spelt out in law) is that an AI is legally no different from a person – any OUTPUT is perfectly CAPABLE of infringing copyright. Cheers, Wol The Law is already Established; Work for Hire Posted Mar 26, 2026 22:55 UTC (Thu) by mb (subscriber, #50428) [Link] (6 responses) Sure. But extremely unlikely. The Law is already Established; Work for Hire Posted Mar 27, 2026 0:36 UTC (Fri) by khim (subscriber, #9252) [Link] (4 responses) Actually it’s extremely likely.OpenAI itself cries foul and tells us that Deepseek “stole their IP” when it trained their models on OpenAI data… but how can it’s protect it’s right to rip of millions of authors of public works on the internet and yet, simultaneously, claim Deepseek couldn’t do what it does?If transformation of pile of data into LLM doesn’t need permission because output of LLM is not “derived work” then how can OpenAI claim Deepseek violated anything… and if such transformation leaves enough traces of original works to be caught up and presented in the court as IP laws violation then how could you be sure your ext4 implementation wouldn’t violate rights of other people in the same way?We have absolutely no idea where the whole thing would end… we just know it would be very, very, VERY ugly. The Law is already Established; Work for Hire Posted Mar 27, 2026 2:22 UTC (Fri) by timrichardson (subscriber, #72836) [Link] (3 responses) A user of Anthropic or OpenAI models does so under a contract; the claims in this case are probably not copyright claims, but breach of contract, I guess. I’m not sure you have a “gotcha”.The Law is already Established; Work for Hire Posted Mar 27, 2026 13:15 UTC (Fri) by gfernandes (subscriber, #119910) [Link] (2 responses) Any contract between the LLM service provider and the user doesn’t indemnify the user (or the service provider) of copyright and patent law. Both need to abide by both laws. And neither can claim copyright (or original ideation) of anything in the training material.The Law is already Established; Work for Hire Posted Mar 27, 2026 15:27 UTC (Fri) by kleptog (subscriber, #1183) [Link] (1 responses) Of course they can, see the Microsoft Copyright Commitment: > . to defend these customers and pay for any adverse judgments if they are sued for copyright infringement for the use of the Azure OpenAI Service outputs. [1] Businesses would never use it unless they were indemnified. It doesn’t help with patents though, but that’s a much lower risk.It’s all civil cases, not criminal, so Microsoft can indemnify it fine. Other commercial models do the same. If you use a free model you’re out of luck, there’s no contract you can rely on. [1] https://blogs.microsoft.com/on-the-issues/2023/09/07/copi.Different uses of indemnify Posted Mar 27, 2026 16:33 UTC (Fri) by farnz (subscriber, #17727) [Link] I think you’re talking past each other; my understanding of what gfernandes meant is that no amount of indemnity language in the AI contracts can ensure that neither AI user nor AI provider is at risk from a lawsuit by a third party copyright holder as a consequence of the use of AI.At best, the indemnity language can choose (as the Microsoft Copyright Commitment does) whether it’s the AI user or AI provider who ends up paying for the resulting lawsuit. There’s no indemnity clause you can write that means that a third party whose legal rights are infringed by your actions cannot win a legal action against at least one of the parties to the contract.The Law is already Established Posted Mar 27, 2026 3:24 UTC (Fri) by ebiederm (subscriber, #35028) [Link] Code generating AIs have been know to regurgitate exact copies of other peoples code. So much so that a filter was built to search through preexisting code and take action if found. So it seems entirely reasonable that if in doubt assume code generated by AIs was cut and pasted from someplace with a license different than you would like and the AI simply did not credit the author.That violates most Free Software Licenses. From time to time people get litigious when their licenses are not followed, so it can be prudent to be careful. Add to that there have several well publicized firings of people at AI companies that were fired for arguing for ethical behavior. So I would not trust the AI companies to have done careful due diligence on this matter. I would rather presume like many start-ups they figure anything they can get away with must be legal and ethical.I hope I am wrong but without doing my own due diligence there is no way I can tell. So I think assuming AI generated code is just copy and pasted from somewhere and not credited is the conservative safe response to AI generated code. License doesn’t need copyright Posted Mar 26, 2026 19:30 UTC (Thu) by roskegg (subscriber, #105) [Link] (2 responses) There is another aspect to the matter; since ownership is assigned to the user of the LLM, they can still attack a license to it.It may or may not be eligible for copyright, but as long as there is some type of ownership claim, an open source license should be valid. Copyright isn’t the only type of intellectual property. There is trademark, trade secret, patent, and probably some others. License doesn’t need copyright Posted Mar 26, 2026 21:51 UTC (Thu) by dvdeug (subscriber, #10998) [Link] (1 responses) I’m big on not using the phrase intellectual property. This is a good example.Open source licenses don’t usually have anything to do with trademarks, which are largely irrelevant to LLM generated code. Trade secrets are antithetical to open source; one you make something generally available, it’s not a trade secret. It seems as unlikely AI generated material can create patentable work as to create copyrightable work, and patents are bureaucratic nightmares; getting one is an expensive complex process.There are other things classed under IP; if you’re creating a computer chip, there’s protection for maskwork, and I believe protection for boat models. They’re not things you randomly wander into, or would license under a standard open source license. If AI generated code isn’t copyrightable, then you basically have no power to demand people follow your rules with it, aka license it.License doesn’t need copyright Posted Mar 26, 2026 22:48 UTC (Thu) by Wol (subscriber, #4433) [Link] > If AI generated code isn’t copyrightable, then you basically have no power to demand people follow your rules with it, aka license it. Well, if it’s AI generated, it can’t be yours to licence! You’re not the creator, you have no rights. Simple as. The question is “is it a derivative of its training material”, aka “did copyright pass THROUGH the AI from input to output?”.If you can show the AI work is a derivative of your training material and prompts, then you put your creative copyrights in, you get them back out the other end. Cheers, Wol “”forking OpenBSD”” might be easier. Posted Mar 26, 2026 23:49 UTC (Thu) by mirabilos (subscriber, #84359) [Link] (4 responses) It isn’t. Signed – the MirBSD developer “”forking OpenBSD”” might be easier. Posted Mar 27, 2026 0:08 UTC (Fri) by rahulsundaram (subscriber, #21946) [Link] (3 responses) > It isn’t.It might be a tad bit more helpful to add some details as to why “”forking OpenBSD”” might be easier. Posted Mar 27, 2026 0:54 UTC (Fri) by mirabilos (subscriber, #84359) [Link] (2 responses) BTDT *shrug* “”forking OpenBSD”” might be easier. Posted Mar 27, 2026 2:03 UTC (Fri) by buck (subscriber, #55985) [Link] (1 responses) maybe you just wanna have an LLM pick up your todo list http://www.mirbsd.org/projects.htm and then that fork will be done *grin* “”forking OpenBSD”” might be easier.Posted Mar 27, 2026 2:33 UTC (Fri) by mirabilos (subscriber, #84359) [Link] I have nothing polite to respond to that, so kindly move your fashtech-related replies elsewhere, thanks.Public domain code (or even compatibly licensed code) does not need a grant of permissions Posted Mar 27, 2026 15:05 UTC (Fri) by jejb (subscriber, #6654) [Link] Specifically this statement: > So as of today, the Copyright system does not have a way for the output of a non-human produced set of files to contain the grant of permissions which the OpenBSD project needs to perform combination and redistribution. Is not necessarily correct.If a contribution is public domain (or compatibly licensed) there’s no need of any additional grant to the project because all the necessary rights are already granted. Thus if it is concluded (as several courts are tending to) that machine produced output that isn’t a derivative work of something else can’t be copyrighted it becomes ipso facto public domain and thus can be incorporated into any project under any licence (public domain being compatible with any form of reuse).Rewrite ZFS with Claude for the lolz Posted Mar 27, 2026 15:52 UTC (Fri) by bluca (subscriber, #118303) [Link] (1 responses) Someone should now pull a reverse UNO card and have Claudebot rewrite ZFS for Linux under GPL2 and merge it into the kernel.Just for shits and giggles, really Rewrite ZFS with Claude for the lolz Posted Mar 27, 2026 15:56 UTC (Fri) by bluca (subscriber, #118303) [Link] Also other than being funny as heck, it would help quickly clarify all the remaining legal doubts, given Oracle would sue after about 4.2 microseconds Copyright © 2026, Eklektix, Inc. Comments and public postings are copyrighted by their creators. Linux is a registered trademark of Linus Torvalds []