Codeberg’s Ban on AI-Generated Code: Vote Margins, Reasoning, and What It Means for Open Source

What Codeberg adopted

Codeberg announced two distinct policies :

1. A binding commitment that Codeberg will not use any project or user data to train LLMs or other generative AI systems.
2. A ban on hosting projects that "mostly consist of" code written by generative AI tools (including services such as Claude and OpenAI Codex). This was codified into Codeberg's Terms of Use .

The ban does not apply to established projects with active communities that occasionally accept LLM-assisted contributions, ordinary projects using AI for small assists, or minor experiments .

Vote margins

The LLM policy was approved by Codeberg's dues-paying active members after a two-week vote :

• 358 in favor (~69%)
• 144 against (~28%)
• 14 abstentions (~3%)

Multiple outlets report these exact numbers . Turnout was roughly half of all active members, and about 71% of those who took a side voted yes .

Reasoning: infrastructure strain, cost increases, and erosion of trust

Codeberg gave several interconnected reasons for the ban :

• Infrastructure and cost strain. The volunteer-run platform cited a sharp increase in hosting costs and resource consumption driven by low-quality AI-generated projects being dumped onto the forge. These projects consume CI/CD minutes, storage, and bandwidth with little to no ongoing human maintenance .
• Erosion of trust in FLOSS. Codeberg argued that LLM-generated code floods the commons with untested, unmaintained, and unaccountable software, degrading the trust that the free/open-source ecosystem relies on for collaboration and security audits .
• Unclear legal and copyright status. Generated code has ambiguous copyright provenance, making it difficult to comply with Codeberg's existing ToU requirements that projects have clear licensing .
• Security and safety risks. The platform noted that AI-generated code lacks adequate safeguards and may introduce hidden vulnerabilities that no human has reviewed .
• Volunteer maintainer burden. The flood of AI-generated pull requests was overwhelming volunteer reviewers, mirroring the problem that motivated Godot's similar ban .

How enforcement will work

Enforcement is designed to be pragmatic but firm :

• Terms of Use restriction. The prohibition is now written into Codeberg's Terms of Use. Projects that "mostly consist of" LLM-generated code are prohibited from being hosted .
• Reporting and moderation. Enforcement relies on community reporting and moderation. Codeberg has not announced automated detection tools; moderators will evaluate projects on a case-by-case basis .
• Gray areas exempted. Small experiments, ordinary projects that accept some LLM contributions, and established active communities are explicitly not targeted . The policy targets "vibe-coded" projects — repositories that are essentially AI-dumped with little to no human curation .
• Parallel data-use ban. The first resolution — that Codeberg will not scrape or use hosted data for AI training — is a binding organizational commitment, enforceable by the membership .

Godot's similar stance (shortly before Codeberg)

The Godot Foundation announced a very similar policy on June 30, 2026 — roughly three weeks before Codeberg's platform-wide vote .

Key elements of Godot's policy :

• Ban on AI-authored code. No substantial code written by AI is accepted into the engine's codebase.
• Ban on autonomous AI pull requests. Pull requests submitted by AI agents are automatically rejected.
• Ban on AI-generated communication. AI-generated text in human-to-human contributor messages is also prohibited.
• Limited exception. AI assistance is only allowed for minor, menial tasks like code completion, regex, find-and-replace — the kind of tooling that merely helps a human write code.
• Stricter PR controls for new contributors. New contributors (defined as those with three or fewer merged pull requests) now need explicit maintainer approval before submitting major features or significant refactoring.

Godot's stated reasoning directly overlapped with Codeberg's: an overwhelming surge in AI-generated pull requests was drowning volunteer reviewers, lowering code quality, and eroding trust in the contribution process .