Codeberg just updated its Terms of Use to ban “vibe-coded” projects. The open-source hosting platform is actively rejecting repositories that consist mostly of code generated by tools like Claude and OpenAI Codex. The restriction passed after 358 members voted in favor against 144 dissenting votes.
The platform officially merged pull request #1253 at its 2026 general assembly to halt the economic drain of automated ghost projects. Codeberg operates as a democratic, member-run alternative to mainstream tech hubs. With nearly 400,000 users, the platform relies entirely on community funds to keep its servers running.
Bulk-generated AI software directly threatened that operating budget. Single developers were using large language models to pump out massive repositories in minutes. Those projects immediately started eating up continuous integration and delivery (CI/CD) testing resources. They also consumed vast amounts of storage without ever building real, active user communities.
The organization established a clear legal boundary to enforce the new rule:
- The Copyright Anchor: Codeberg grounded the new Rule 7 in German copyright law, specifically citing sections § 2 (1) 1 and § 2 (1) 3.
- The Vibe-Coded Ban: Repositories generated in bulk by LLMs and uploaded directly to the service are now strictly prohibited.
- The Human Exception: Developers can still use AI-assisted coding tools for individual functions or snippets, provided human oversight remains the primary driver of the project.
By tying the ban to strict copyright requirements instead of subjective complaints about code quality, Codeberg’s board created a defensible legal shield. This protects the non-profit from future ownership disputes involving AI training data.
GitHub Embraces AI While Codeberg Protects Its Infrastructure
This vote exposes a hard split in how open-source platforms approach artificial intelligence. Microsoft-owned GitHub actively integrates tools like Copilot into every layer of its developer experience. Codeberg is now moving in the exact opposite direction. The platform is aligning itself with other open-source entities like Godot and Flathub that also push back against accepting LLM-generated software.
The decision proves that unfiltered AI code generation carries real physical costs for independent platforms. When generative tools allow one person to create hundreds of applications effortlessly, the hosting provider pays the server bill to store and test that code. Codeberg members simply decided they no longer want to fund the infrastructure for empty AI projects.