It’s a tough spot, Codeberg has found themselves in and I wish them luck. But beyond that, this is (yet) another reminder that in 2026, if you don’t self host it, the cloud is just someone else computer
From what I understand from their blogpost, it was mostly about conserving their own resources, because a lot of the vibecoded projects they got were uploading insane amounts of binary releases and wasting resources on CI/CD while having no users or other collaborators.
They wanna reserve more server space for projects, that actually productively use it, like bigger FOSS projects with actual users and contributors.
That is perfectly understandable for me, even without taking my huge distaste for AI into consideration. Anyone that thinks, that this small, community funded project is obligated to host their huge slop repos semms pretty entitled to me.
before reading the blog post I was thinking the same. now I don’t.
the worst of the LLM projects have no place on codeberg that’s for sure, but there would have been better ways than a blanket ban to limit the resource consumption of LLM and crypto projects. codeberg already has a storage quota system, they could be giving a lower quota for LLM projects, maybe also disable free CI for them, which I am a bit surprised they have given. and solve the reputation problem with banners. but no, total blanket ban based on feelings, it is.
There is simply no way that is true. First, the legal arguments are dodgy:
There is a good chance that LLMs are sufficiently transformative that courts will decide they don’t infringe the copyright of their sources.
Even if not, hosts will get DMCA-style safe harbour protection and the most they’ll be liable for is takedown requests, which is a major thing for any large host.
Second, there is no fucking way a western court is going to tell every big tech company they have to delete 99% of the code that was written since the start of the year, even if the law as written literally said verbatim, “use of any LLM output for any purpose is breach of copyright” because it doesn’t take a conspiracy theorist to realise that it’s politically impossible.
You may not like that, but it means that, again, Codeberg is doing this based on feels.
You are missing the point entirely. LLMs regularly generate code that is a near verbatim copy of existing copyrighted code, but with almost no way for the LLM using person to notice that. LLMs being sufficiently transformative might be an argument about the use of training material, making the resulting model not a copyright violation itself, and thus might protect the companies that produce and offer these models, but it says nothing about the actual output of a model.
It is only a question of time before some enterprising law firm decides to mass scan open-source projects and weaponize their findings similar to patent trolls or file-sharing legal threats. This has a long history in Germany where Codeberg is located, and even if a court rules that the host itself is only responsible for removing such copyright violating code, it will require significant effort to do so with constant legal fights as the attacking lawyers will try to figure out the identity of the person responsible so that they can blackmail them with cease and desist legal fees.
Politics will not care about some hobbyist open-source projects and large companies will spend a lot of effort to obfuscate their code to prevent this legal trolling to affect them.
People can upload code that violates copyright too. No llm required.
So if that is the concern have a rule about not violating copyright (which they may already have). And there is likely already a process for handling that.
This is the problem I have with the situation: the solution chosen (banning llm code) doesn’t really address the concerns they’ve raised directly. It’s just apologetics to make it sound like it’s reason-based rather than “we don’t like it”.
It’s their platform, they can do what they want, but let’s not pretend it’s not code-puritanism.
Yes, they can upload copyrighted stuff, but people typically don’t do so on large scale and when they knowingly do it these days they typically try to hide their tracks well enough that law firms know it will not be a lucrative business to try and blackmail them.
And one of Codeberg’s main points is the unknown copyright status, you just failed to understand it.
LLMs cause a lot of people to unknowingly violate copyright, and those people then become easy targets for malicious copyright litigation. And Codeberg is caught in the middle of that and doesn’t want to be involved in the resulting mass legal cases because they are just a small volunteer organisation without a legal team.
And Codeberg’s explanation is very clear that is isn’t a blanket ban on LLM generated code for “purity” reasons. It is a risk mitigation strategy against projects that are mostly LLM generated.
What hasn’t happened? Sure, this isn’t wide spread yet, but it also took malicious law firms a while before they realized going after p2p torrent users is a lucrative business.
And I seem to have a much better grasp at copyright than you do.
P2P torrent users are nothing like a code hosting platform. 🤣
As has been said - their liability for llm code is no different from hosting any other code. They already have the same risk. And that is typically that they must respond to take-down notices which they are already doing.
They are already dealing with all of the problems they would be dealing with with llm code.
It’s just code puritanism wrapped in pseudo legal justification. Own it.
It is a totally different thing to have the occasional copyright take down request from a legitimate copyright holder, or hosting code that is in the majority likely copyright infringing and just waiting for someone to start targeting for mass copyright litigation.
It is not. There are plenty of studies showing that LLMs spit out code that is near verbatim to existing code and LLM companies even go so far as to instruct their models to not also add the corresponding license/copyright headers with that code.
There are probably law firms analysing common code patterns LLMs often use right now and are approaching copyright holders of similar enough code to buy up the rights. It might not all stand up in court, but it will be enough to scare some people into settling for fee that guarantees a profit for these law firms. This is a tried and true method for an entire industry of law firms.
You realized that those parts refer to two different things? But since you clearly live in a world entirely detached from reality I don’t think this makes sense to discuss further.
You know what’s funny? Codeberg has said nothing about being concerned about liability.
Their terms service change simply mentions that they require certain licenses and that they have concern over the licensing of LLM meeting that standard.
Event their follow up communication says absolutely nothing of liability.
LLMs cause a lot of people to unknowingly violate copyright, and those people then become easy targets for malicious copyright litigation
I’d buy that. As a mediocre musician, one can even subconsciously ‘sample’ a copyrighted piece. I mean, there are only so many chords and combinations, melodies, etc, so duplicates, and even exact duplicates are bound to happen whether intentional or not. If you upload your track to say SoundCloud on the professional plan with which you intend to generate revenue, the track is heavily scrutinized, and often rejected for this very reason. You actually have to prove it’s OC by various means. They don’t want any pieces parts of any litigation that may even mildly involve them. So, I can see that.
AIs already pillaged all the code in the world.
This is fighting a lost fight. We won’t manage to force big AI firms that literally props a country to stay afloat to make their AI “forget” or “untrained”.
Regulators let the pillaging happen so now AI does know how to code. They dont imitate they truly code.
But IMO the real fight is about things like malus that are the real dangerous and malevolent actor here. They will take your open source code and resell it to big businesses.
And you are also misunderstanding my point. Yes the cat is out of the bag and companies will absolutely copyright wash their codebases like that, but they have big legal teams to defend against copyright trolls and generally do not publish most of their code base for anyone to see.
Small hobbyist open-source projects that make up near 100% of the projects hosted on Codeberg on the other hand are easy marks for malicious litigation, just like home users torrenting movies were in the 1990ties and early 2000.
You’re right I wasn’t really thinking about the possibility of verbatim reproduction. I guess we’ll see about that.
Codeberg will need efficient procedures to handle IP trolls anyway, getting a bunch more takedown requests ought not to make a big difference. And if somehow it does start to be a problem… that would be the time to blanket ban or take similar drastic action. If it’s possible to detect now, it will be possible to detect then.
Politics will absolutely care about having a precedent which leaves big tech exposed. Any case against little fish will be jumped on by Zuck and Bezos before you can say “datacenter”.
No, these malicious law firms intentionally go after the small fish, who are much more likely to be intimidated and give in to a settlement payment before it even reaches the courts.
And for Codeberg it makes absolutly sense to minimize the issue for them now before it becomes an even bigger issue.
And all 3 hinge on the premise of courts will do ‘sane’ things, without any kind of justification for them. So yeah, all 3 are moot if you remove that assumption.
@FishFace Nothing is "politically impossible," and AI is not inevitable.
When rich people get out of hand, there is always the same correction that takes place, sooner or later. Right up until the very moment it happens, people with varying degrees of comfort in the existing regime are squawking that the world will end if we mess with the Load-Bearing Pedophiles.
They have less chance of making their tyranny permanent this time than at any previous time it has been tried.
Prepare your mind for the idea that you and everyone you know might have to personally pick up a shovel and dig, before the mess they are making right now gets cleaned up.
I think your feelings are irrelevant to the admins’ and don’t change the fact that they aren’t being honest (with people, or quite possibly themselves)
A) I haven’t tried to decide anything. I’ve commented on their public statement, with reasons.
B) since you don’t seem to have understood what I’m saying (see above, and previous comment) you’ve not given me any reason to care about your assessment of me.
You’re not interested in discussing my comment… just stop replying instead of calling me names?
Not true, this does not occur frequently. This study and software LiCoEval from 2024 found 0.88% to 2.01% of code “strikingly similar to existing open-source implementations”. Afaik this is mostly textbook examples, snippets from stack overflow snippets or common github repositories, often replicated api examples and language boilerplate. How you prompt and refine also matters, and for generating novel code or business logic the LLM simply cannot use memorized snippets.
Presumably since then LLMs have worked to reduce that number of memorized code. Since LLMs cannot memorize all their training data, that number is limited. LiCoEval can find the often memorized examples and train to remove them, or suppress them, or they find other ways to reduce direct reproduction from memorization. For example it would be possible to do what malus.sh does with all the training data. Then it cannot memorize copyrighted code.
So for a model that came out 2026 this already small number might not be that relevant anymore.
Copyright can also apply to code rewritten in another programming language, if the overall structure and functionality is identical. Although tbh, that is a bit of a grey area, and in previous rulings there is some precedence that if the “translation” involves significant human effort then it can be considered to be sufficiently transformative. But promting an LLM to “rewrite this in Go” very likely doesn’t pass that relatively high bar.
The study you linked is interesting (and likely underestimates the issue*), but it also shows that it is possible to automate detection of copyright infringing code in LLM output, and even just 2% is more than enough to open a case for malicious litigation. Large companies can write that off as the cost of doing business, but for small individual open-source developers this is a significant risk.
‘*’ Because human programmers usually reuse existing licensed code as libraries, while LLMs are prone to rewrite them, and as explained above a rewrite is at an elevated risk of being considered copyright infringing.
You can also set a LLMs to completely rewriting code snippets from scratch to perform the same function, writing and fulfilling the same unit tests. Not just rewrite in a different language or changing variable names. Basically the transformative abilities of LLMs can make copyright obsolete.
Basically the transformative abilities of LLMs can make copyright obsolete.
That is wishful thinking. Copyright was originally designed to protect human creativity, and a lot of the laws and precedence cases involve human effort specifically to distinguish between plagiats and original works.
As long as courts don’t start to rule otherwise, LLM generated code is not considered to be copyrightable itself, thus it can not be original work as defined under copyright laws. This in turn lowers the threashold for such code to be considered a derivative regardless of if there is a clear pathway from one to the other. All it takes is someone to claim that it is derivative, and the burden of proof that it isn’t largly falls onto the person accused. And if the code is functionally identical and with no significant human involvement, that is very hard to proove.
The IP law is purely a function of economic power, so we know how that will go. To the maximum benefit of the plutocrats that rule our global civilization.
But the more interesting part is the philosophical discovery: If a piece of code can be generated from scratch by a machine that is not creative, then any comparable piece of code is also not creative. In other words, only the very first time a new type of code or algorithm is written can there be creativity - if one follows your argument that LLM machines cannot be creative. This essentially raises the bar for copyright, and makes an objective test of what is copyrightable or not possible. If you can describe something in a prompt, and get a result of comparable function and quality to a human written version, that code is not creative either, therefor not copyrightable.
Obviously we’re not quite there yet, but the more LLMs improve, the less code becomes copyrightable.
Coding becomes pure craftsmanship, like cutting and sanding some lumber and nailing it together to build a shed.
So if we were a rational and logical species, the overwhelming amount of copyright would be deleted and become public domain. Because it is now worth only the cheap solar power needed to generate it.
An important consideration is that the usage of “AI” aka currently using like LLMs is viral, in a worse sense than eg.: the GPL is viral, and it’s completely understandable that some hosters would prefer to avoid the community problems that that causes.
As soon as you set quotas, you get assholes suddenly exploiting them to the max, who didn’t before when it was “unlimited”. Anyone involved in hosting services sees this happen time and time again.
We’re not talking about bad code, we’re talking about slop code. It’s different. It’s easy just to generate slop code, it is not so easy to code all by yourself.
