Asahi Linux bans generative AI, DHH says upstreaming is rejected
A maintainer policy against LLM-written code just turned into a public fight with two of the loudest builders in open source.
Asahi Linux, the project porting Linux to Apple Silicon, has broadly forbidden the use of generative AI in contributions. DHH posted the policy text with an eye roll emoji and nothing else. The quoted passage says LLMs harm "software freedom, the open-source community, the open Internet, the environment, digital privacy, user consent, and software developers themselves".
The part that matters commercially came in a follow-up. DHH said his team has been expanding Asahi's work with TouchID, M3 GPU support, DisplayPort over USB-C, deep sleep and MacBook Neo, and that thanks to agent acceleration, Asahi now rejects upstreaming those changes. He added that Asahi moving to GPLv2 changes nothing because the Linux kernel is already GPLv2, that all source changes against the kernel are made available, and that Omarchy itself is MIT.
Steinberger calls it silly
Peter Steinberger, posting as @steipete, went after the logic rather than the licence. He pointed out that the Asahi project was about taking something closed and restricted and opening it up, so adding new restrictions that make it harder for other projects to build on that work cuts against the point. His verdict on the AI clause specifically was that it is just silly.
Neither post includes Asahi's side beyond the quoted policy, and nobody in the sources says how much code was submitted or what exactly was turned down. So this is two builders describing a rejection, not a documented account of the review process.
The business read
The interesting thing here is not whether LLM code is any good. It is what happens to the plumbing when a maintainer says no. DHH's answer is that the downstream work carries on anyway, published under MIT, with kernel changes available as GPLv2 requires. The port gets built either way, it just stops flowing back up.
For anyone shipping on top of open-source infrastructure, that is the practical risk to watch. If a project you depend on bans agent-written contributions, your patches live in your fork forever, and you carry the merge cost every time upstream moves. That is survivable for a team with agent throughput to spare, which is roughly DHH's point. It is less survivable for a solo founder maintaining a patch set by hand.
Nobody knows yet whether other maintainers follow Asahi here. One policy is one policy. But it is the first clean example of an AI contribution ban producing a visible fork-instead-of-upstream outcome, and the people on the other side of it are not quiet about it.
And that AI rejection is just silly.
thanks to agent acceleration, Asahi rejects upstreaming
Steal the numbers, not the vibes
One email a day. Under three minutes. Real MRR, real pricing, real launches from founders building with AI.

