Codeberg Divides Over Generative AI Rules and Open Source Future
Codeberg recently updated its terms to exclude projects largely written by generative AI. While I support democratic governance, I worry this vague policy creates unpredictability for legal Open Source software. Instead of a clear ban or specific rules against spam, the current middle ground relies too much on interpretation. This split over LLMs risks dividing the Open Source community when we should be engaging with the future of software creation together.
I wish Codeberg were more forward-looking here: willing to host the Open Source software of tomorrow, not only software made in the ways its community approves of today.
- free_bip
I think this move by Codeberg is lowkey genius. They're trying to build a community based on a set of shared values. Those who disagree with those values are now self-selecting out of the Codeberg community, which is exactly the point of passing these kinds of resolutions.
The fact that the rule is open for interpretation means that they can effectively shut down malicious actors (i.e those who intentionally misrepresent the degree of AI assistance in writing their code) before they become a problem. It becomes an exercise in judgement instead of a potentially exploitable/loophole-able set of hard and fast rules.
There's no issue with this. This isn't the supreme court. Your income is not on the line. It's a code hosting repository. Go to GitHub if it bothers you.
Only time will tell if this set of values holds up. Personally, I think they will, and that the rules will end up being less strict than people are making them out to be, and that relatively few projects will be removed from the platform.
- jjouett
Its a choice they're totally allowed to make, I just wish we'd get an updated mission statement[1]:
"The mission of Codeberg e.V. is to build and maintain a free collaboration platform for creating, archiving, and preserving code and to document its development process."
Amended to include at the end ", as long as you use the tools we find agreeable."
[1]: https://docs.codeberg.org/getting-started/what-is-codeberg/
- miki123211
This is the problem with choosing platform for moral — instead of practical — reasons.
Platforms that are known for making moral judgments will likely keep making moral judgments in the future. Eventually, they'll make a judgment that you heavily disagree with.
This is why communities founded on ideology tend to shatter, while those founded on a shared goal tend to survive. Linus doesn't care whether you're a GPU expert at AMD, a 16-year-old in Nairobi, or a nazi with a swastika tattooed on your arm. He cares about the quality of your kernel patches.
- PaulDavisThe1st
I have no particular problem with this decision/policy on the part of Codeberg.
I do have an issue with their "tainted code" project that labels various FLOSS projects as "AI tainted".
ardour.org has the follow AI policy statement:
---------
Using LLMs to generate code for Ardour
A word about licensing, copyright and LLMs ("AI"). We require all code contributed to Ardour to be licensed under the GPL (you may retain copyright, but the code must use the GPL). Currently, following Thaler vs. Perlmutter in March 2025, no LLM-generated code may be copyrighted in the USA. The GPL relies on copyright, and cannot function if copyright is not held. Consequently, if you use an LLM (Claude, Codex, CoPilot etc. etc) to create code, it cannot be included in Ardour.
If you want to use LLMs to explore and understand some of Ardour's source code, feel free. But we cannot accept PRs or patches or contributions that are not licensed under the GPL, and if you choose to be honest about it, this means that no LLM-generated code can be accepted by us.
---------
Yet this codeberg-adjacent project considers this project to be AI-tainted. If you're going to do something, do it right or don't do it at all.
- klustregrif
> I could not reliably assign authorship percentages to many of my own recent projects.
This feels like a strange statement. Unless you are going out of you way to disguise AI contributions it's extremely simple to do some git magic to get an exact percentage. But you can also just ask AI and it'll tell you directly. And even if you've tried to disguise it I'm fairly certain that if you set aside 10$ for tokens, it'll be able to deduce how much is AI even with someone having tried to disguise it.
- kccqzy
I don’t even mind forges taking a stance on AI. That’s fine. But I agree with Armin Ronacher here: banning projects that “mostly consist of code written by generative AI tools” is like the infamous Supreme Court “I know it when I see it” line. Yes of course I know a project mostly written by generative AI when I see it.
- CodesInChaos
I think introducing unpredictable new prohibitions is a much bigger problem than the specific bans on cryptocurrency or AI written code. Especially when it leads to existing code becoming unavailable and doesn't just ban new contributions violating the rules.
- neogodless
Related:
https://news.ycombinator.com/item?id=49015635 Protecting our FLOSS commons from LLMs (blog.codeberg.org) 1 day ago, 144 comments
https://news.ycombinator.com/item?id=49003386 Codeberg bans vibe coded projects (codeberg.org/codeberg) 2 days ago, 279 comments