Open Source Says No lists 22 projects banning AI-written patches: Reveneau

Why it mattersA team sending AI-written patches to a project may find the project refuses them.

A new site, Open Source Says No, lists 22 open source projects that will not accept AI-written patches. The site reached the front page of Hacker News on Tuesday. It is kept on GitHub, so anyone can suggest another project.

What the site is

The site is a one-page catalog. Each project is a card. The card names the project. It quotes the exact rule from the contribution guide. It links the file where the rule lives. A person called jlengrand runs it. The source list is kept in a public GitHub repository, jlengrand/oss-banning-ai. Anyone can suggest a project by opening a pull request on that repo.

The 22 projects are a mix of old and new. Some are very large. Zig, the programming language, is on the list. So is QEMU, the hardware emulator. So is Gentoo Linux. Others are smaller libraries and tools, such as Flask-SocketIO, pdfcpu, libvips and okhttp. The list also includes a Rust game engine called Fyrox. It names the photo backup tool immich, the search engine meilisearch and the game framework MonoGame.

What the rules actually say

The wording varies from one project to the next, and that is the point of the site. Some projects ban "LLM-generated content" of any kind. Some ban "generative AI" in code, tests, documentation, images and comments. Some go further. The site quotes stb as rejecting both AI-generated code and work that was "AI-analyzed". Veloren treats any text that passed through a large language model as "tainted". Gentoo's rule names the technology by its research label, "NLP AI".

A few rules are only a request. Immich says maintainers would rather contributors did not send AI-made pull requests, without a hard ban. Most are a straight refusal.

What a team should take from it

A developer using Claude Code, Cursor or Codex can still contribute to most of the open source they depend on. The list is 22 projects, not 220. But the list shows a real pattern. Most of the projects write code that is the shared foundation of other software. An operating system. An emulator. An HTTP library. A compiler. A graphics library. A maintainer there has to read every line that comes in. If the patch author did not understand it, the maintainer still has to debug it.

For a team that is about to send patches upstream, the cheapest check is also the first. Open the project's CONTRIBUTING file. Read the rule. The fastest way to have a patch rejected is to not know the rule is there.

The site is updated by hand and will miss projects. Its maintainer invites suggestions. A team sending agent-written patches to many open source projects should expect this list to grow.

Source

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.