www.it-connect.fr
Codeberg Moves to Ban AI-Generated “Vibe Coded” Projects
358 votes for, 144 against, 14 abstentions. That is the result of the vote that has just pushed Codeberg, the German non-profit Git forge often presented as the ethical alternative to GitHub, into a new direction. Two motions on artificial intelligence were adopted by its members: the first commits the platform to never train a model on your code, while the second simply bans projects that are mostly generated by AI. Here is what we know.
Table of Contents
Two motions adopted, only one sparks debate
Codeberg e.V.'s annual meeting, the Berlin-based association that runs the code hosting platform, ended with a 14-day asynchronous vote. It concluded on July 22, 2026, and the results were published the following day on the official blog by Bastian Greshake Tzovaras, Otto Richter, and William Zijl.
The first motion passed without any difficulty. That is not surprising, since it formalizes the position already described in Codeberg's privacy policy: “The Codeberg forge and its associated services do not use and will not use project and user code or data to train 'artificial intelligence' tools”, the adopted text states. Members also believe these technologies are incompatible with responsible creation and maintenance of free software.
The second motion, however, split opinion, despite a very clear vote. Adopted by 358 votes to 144 with 14 abstentions, and with around 50% of active members participating, it adds a clause to the terms of use. And that clause is not neutral: it aims to crack down on vibe coding. In practical terms, users must not share projects made up mostly of code written by generative AI tools. Claude and OpenAI Codex, two of the most widely used services, are specifically mentioned. Concerns around copyright partly explain this decision, but economics are also a factor.
From AI crawlers to €3,700 SSDs
Codeberg's argument is first and foremost economic, and it comes with numbers. The association explains that a disk model bought for around 700 euros a few years ago now sells for about 3,700 euros, and is often out of stock. Because Codeberg owns its own hardware rather than renting capacity in the cloud, it directly absorbs these price increases driven by demand from AI giants.
There is also pressure from crawlers. According to the article published by the association behind Codeberg, these crawlers do not stop at a git clone: they scan every variant of issue filters, every point in the Git history, including identical files from one revision to the next. The result is expensive database queries and defensive mechanisms that, in turn, degrade the experience of legitimate users. And this is unlikely to stop any time soon: the trend is upward, since web traffic generated by AI agents is becoming increasingly significant.
A third point raised is ghost projects. Codeberg describes repositories run by a single developer and with almost no users, yet which consume as much storage and CI/CD as the forge's largest community projects.
Add to that the usual workload issue: maintainers are overwhelmed by AI-generated contributions, often well-intentioned but costly to review. It is a situation that echoes tensions that recently emerged around commits co-written with Claude on the Rsync project.
"In the case of projects under a 'copyleft' license, LLMs also lead to a phenomenon of 'license laundering', in which code under a 'copyleft' license is stripped of its reciprocity requirements by being 'generated' from training data.", Codeberg explains.
No mass purge, but clear red lines
Codeberg is making it clear: there will be no mass deletion, and no exhaustive list of repositories to remove. These new rules will be enforced on a case-by-case basis, without relying on automated content analysis. Projects with an active community, those that predate LLMs, or those that unknowingly accepted an AI-generated contribution should not be targeted. In other words, Codeberg is expected to use common sense.
However, here are the projects that are no longer welcome on Codeberg:
- Projects autonomously created by AI agents
- Projects written and maintained with heavy reliance on LLMs (vibe coding)
- Projects whose storage or CI/CD usage far exceeds what their contributors could have produced manually
- AI-generated tools designed to facilitate the use of AI
- AI-generated contributions submitted in violation of a project's policy
This is about Codeberg, but it is a common problem for all forges. Recently, I published an article about the saturation seen on GitHub: the volume of AI-generated commits is pushing Microsoft to rent capacity from AWS.
