Codeberg declares war on “vibe coding”: two resolutions against LLMs

Codeberg e. V., the non-profit organization behind the free-software forge of the same name, put two resolutions on large language models (LLMs) to a vote of its members. Both passed after a 14-day ballot with a turnout of roughly half the active members.

The first resolution enshrines a principle already embedded in the privacy policy: users' code and data have not been and will not be used to train generative AI. In the organization's view, such technologies are incompatible with responsible open-source software development.

The second, more contentious resolution (358 votes in favor against 144 opposed, with 14 abstentions) adds a ban on “vibe-coded” projects to the usage rules — that is, on software written predominantly by a machine.

Everyone pays for hungry models

The authors offer a whole range of arguments. AI companies' crawlers methodically read through every page of the forge — every combination of issue filters, Git history, files at any point in history, even ones that never changed. These pointless requests overload the databases and force administrators to spend effort on protective mechanisms that hit ordinary users too. Meanwhile, cloning a repository via git clone remains freely available.

Hardware is a separate pain point. A drive that cost 700 euros a few years ago now runs 3,700 and is often out of stock: demand for SSDs and memory from data centers has driven prices sky-high. Codeberg owns its servers, but expanding the fleet has become noticeably more expensive. The organization also points to the growing digital divide, the rising cost of electricity and water near data centers, and lobbying for looser environmental regulation: in Frankfurt, data centers already consume 40% of the local electricity.

A team of one

Special attention is paid to the social side. A “vibe coder,” in the authors' view, works as if a large team stood behind him — when in reality he is alone with a “statistical machine that turns energy into code.” Such projects breed active commit histories, heavy CI/CD runs, and bulky releases while having almost no users at all. Spending donations on hosting such “ghost” repositories is something the organization considers unwise.

More serious still is the erosion of trust. Maintainers are drowning in well-meaning but low-quality generated changes, suspicion is on the rise, and copyleft code stripped of its reciprocity requirements during “generation” — so-called license laundering.

How it will work

There will be no mass deletion of repositories. Side experiments, narrowly specialized scripts, and projects with a history predating the LLM era will not be affected. But projects created autonomously by AI agents, repositories with a disproportionately large resource footprint, and tools entirely tied to the LLM ecosystem may turn out to be unwelcome guests. Codeberg emphasizes: this is a venue for human collaboration, not a dumping ground for throwaway machine code.

Source: Lobsters