Codeberg prohibits AI training and AI mass projects
The open-source software platform Codeberg will no longer allow its hosted projects to be used for training artificial intelligence models. The platform also plans to restrict projects largely generated by AI.
Codeberg, a non-profit, community-operated platform for hosting free and open-source software, has prohibited the use of its hosted code and data for training large language models (LLMs). The decision stems from concerns over AI's increasing infrastructure demands and energy consumption.
The Berlin-based association, which offers Git hosting based on Forgejo alongside CI/CD and Pages features, announced changes to its terms of use. Codeberg considers projects largely created by AI, with minimal human oversight, problematic. These so-called "vibe-coded" projects, where content is primarily generated by language models without long-term human maintenance or development, may soon violate the platform's rules.
Codeberg also cites practical reasons for its stance. Automated data retrieval by AI services burdens the platform's servers and databases. Specifically, the automatic fetching of numerous views, filters, and history states strains the infrastructure, necessitating restrictive measures by administrators that can impact legitimate users.
The members' meeting approved the motions with 358 votes in favor, 144 against, and 14 abstentions. However, the association has not announced immediate automated searches or mass deletions. Projects with active communities or established development histories predating the LLM boom are unlikely to be affected. Codeberg remains critical of projects autonomously generated by AI agents or where development is heavily reliant on language models and consumes disproportionate resources.