Codeberg bans LLM training and vibe-coded projects in member vote
Codeberg, the German non-profit that runs a GitHub alternative code forge, just made two concrete policy decisions about generative AI. Both passed in a member vote after a 14-day asynchronous window following the organisation’s annual assembly.
The Two Votes
The first motion formalises a position Codeberg had already held informally. The organisation will not use code or data from its users and projects to train AI tools, including LLMs. The exact language from the vote:
“The Codeberg forge and its associated services are not and will not use the code or data of projects and users to train ‘Artificial Intelligence’ tools such as Large Language Models, whose purpose is to create output modelled after their training input.”
That language was already echoed in Codeberg’s privacy policy. The vote turns it into an official association position rather than a clause in legal documentation.
The second motion changes Codeberg’s terms of use to restrict what the organisation calls vibe-coded projects: software built largely or entirely through LLM-generated code with minimal human authorship or oversight. This one was more contested. Around half of active members voted. The result was 358 in favour, 144 against, and 14 abstentions. That’s nearly three-quarters support, but 144 members voting against it shows this wasn’t a settled consensus.
Why the AI Crawler Problem Drove This
Codeberg ties both votes to a real infrastructure cost. AI company crawlers have been hammering the platform to scrape code for training data. The issue isn’t access: Codeberg already allows open Git cloning with no special permission required. The problem is that crawlers ignore the Git mechanism and instead load every individual page on the site, including every issue filter variant, full Git history, and file states at every historical commit, even unchanged ones.
That generates database query volume the platform wasn’t built to absorb. Codeberg says its system administrators have had to spend time building defensive measures, including rate limits and access blocks, rather than working on actual infrastructure improvements. Those same defences also degrade service for legitimate users.
Codeberg’s broader argument is that the infrastructure cost of running LLMs doesn’t stay with the companies that build them. Higher hardware demand, energy draw, and environmental impact get pushed onto platforms like Codeberg that never signed up for any of it.
The Vibe-Coding Concern
The reasoning behind the vibe-coding restriction is more speculative, and Codeberg is upfront about that. Maintainers say they’ve noticed a pattern: a single developer behind a repository, yet the project runs CI/CD pipelines sized for a team, ships release binaries on a frequent and heavy schedule, and claims support across more platforms than its user count would justify.
Codeberg isn’t saying LLM-assisted code is automatically a problem. The concern is that one person prompting a model can produce a project that looks like it has a team and user base behind it, drawing on server resources and storage as if that community actually existed. The blog post does not back this up with repository counts, storage figures, or CI/CD minute totals. Whether this pattern is widespread or a handful of visible cases getting generalised is not substantiated.
What’s Still Unclear
Enforcement details haven’t been published yet. Codeberg acknowledges it owes the community more clarity on how the new terms will be applied, and says it will share that separately. For any maintainer currently hosting a project built substantially through LLM assistance, that follow-up post is the one to watch. It will determine what “vibe-coded” means in practice and what happens to projects that fall under the definition.
For context: GitHub has Copilot built directly into its product and trains on public repository data as a matter of course. Codeberg is now formally defined by rejecting that model rather than treating AI capability as a feature.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
