Codeberg Members Vote to Ban 'Vibe-Coded' Projects and Bar LLM Training on Its Code

Overview

Members of Codeberg e.V., the nonprofit behind the volunteer-run Codeberg code-hosting platform, have voted to prohibit projects made up mostly of AI-generated code and to formally rule out ever using hosted code or user data to train AI systems. The decisions, detailed in a blog post published by Codeberg on July 23, follow two member votes, the more contested of which passed 358 to 144 with 14 abstentions.

What We Know

Codeberg’s membership held two separate votes on artificial intelligence policy. The first addressed data use, reaffirming that “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,” according to the post, co-authored by Bastian Greshake Tzovaras, Otto Richter, and William Zijl.

The second vote, which the authors describe as “more controversial,” passed on a tally of 358 votes in favor, 144 against, and 14 abstentions, with turnout of around 50 percent of active members. It directs Codeberg to change its terms of use to prohibit what the post calls ‘vibe-coded projects.’

That change is now live. Codeberg’s Terms of Use now state, in section 2(1)7, that “you must not share projects that mostly consist of code written by ‘generative AI’-tools (including services such as Claude, OpenAI Codex).”

Codeberg’s blog post frames the decision around two pressures: cost and community trust. On cost, the organization says the price of the hard drives it sources for its infrastructure has climbed from roughly €700 to about €3,700 within a few years, and the drives are often out of stock. On trust, the authors argue that heavy LLM use in free and open-source software is becoming “a multidimensional attack on the trust between contributors and the very idea of convivial collaboration itself.” They also warn that AI code generation enables “license laundering, where copyleft code is stripped of its reciprocity requirements by ‘generating’ it out of the training data,” and object to spending “precious donation money into hosting of large ghost projects” with no real community behind them.

Separately, Codeberg’s membership also voted to ban cryptocurrency projects from the platform, citing a similar move by SourceHut in 2023, and passed a two-thirds-majority amendment adding a commitment to “oppose discrimination and promote a diverse FOSS community” to Codeberg’s stated purpose, though that amendment has not yet been merged.

Not everyone in the open-source community welcomed the vote. Armin Ronacher, creator of Flask and co-founder of AI agent company Earendil, wrote on X that “I think this is a very bad move, and the people behind Codeberg should re-consider their stance.”

What We Don’t Know

Codeberg’s post does not spell out how the vibe-coding ban will be enforced. The Register reports that active policing looks unlikely unless a project draws outside attention, given the nonprofit’s limited resources. It is also unclear how many hosted projects currently run afoul of the new clause, since Codeberg has not published an estimate, and the diversity and anti-discrimination amendment, while it cleared a two-thirds vote, remains unmerged, so its final wording could still change.

Analysis

The vote places Codeberg among a growing number of open-source venues pushing back on AI-generated contributions after months of friction over low-quality automated pull requests and security reports, a trend previously reported by The Machine Herald — coverage that also noted Gentoo’s earlier migration to Codeberg itself over GitHub Copilot prompts. Where that earlier wave of countermeasures largely targeted the quality of AI-assisted contributions, Codeberg’s new terms go further: they bar AI-majority projects outright and foreclose using the platform’s archives as AI training data altogether.