Asahi Linux is the project that has brought Linux to Macs with Apple Silicon, and on November 12, 2026 it emerged that it is migrating all its repositories from the MIT license to GPL v2 or later. The reason given by the team is specific: someone was taking code that had not yet been published and rewriting it with AI models to hide where it came from.
The choice of license is not an administrative detail. MIT allows someone else's code to be taken into a closed product as long as the copyright notice is kept. GPLv2 requires publishing the source code of modified versions that are distributed, and that obligation is what turns an opaque rewrite into a legal problem for whoever does it.
There are two caveats before going further. The first is that the project's official announcement, with the exact list of repositories and the effective date, has not been published as of today, November 13, 2026: the change is known from the weekly roundup of November 12. The second is that GPLv2 does not prohibit using AI to program: it changes the cost of hiding the origin of the code and adds obligations for anyone distributing a modified version.
Leaving MIT to adopt GPLv2
The MIT license is permissive by design. Anyone can use, copy, modify and sell the code, even inside a proprietary product, with the sole condition of keeping the copyright text and the permission notice. It is a common license in development tools because it makes adoption easy and requires nothing in return.
That freedom has a less pleasant side for a reverse-engineering project. Asahi Linux writes drivers for hardware that Apple does not document, and it does so openly with the idea that the work should serve everyone. If someone can take that code, close it off and give nothing back, the community pays for the work and someone else reaps the benefit.
| Aspect |
MIT license |
GPLv2 or later |
| Using the code in a closed product |
Yes, keeping the copyright notice |
No, the distributed version remains under GPLv2 |
| Publishing modifications |
Not required |
Required if you distribute the binary |
| Attributing authorship in the binary |
The copyright notice is enough |
The notices are kept and the code is added |
| AI rewrite presented as one's own |
Falls outside the project's scope |
Must be published if the result is a derivative work |
The fourth row of the table is what explains the change. A rewrite generated by a model can produce code that is functionally equivalent and shows no trace of the original, and under MIT that is enough to publish it as one's own. With copyleft, the distributed version carries the license and requires the source to be shown.
Copyleft as a deterrent
It is worth being precise about what GPLv2 does and does not do. It does not prevent training models on GPL code, it does not prohibit a program from generating similar code and it does not pursue anyone who copies for internal use. What it does prevent is distributing a modified version without publishing its code under the same license.
For the case at hand, the practical difference is enormous. A competitor that rewrites Asahi drivers with the help of AI and puts them into a closed product commits an infringement if the result is a derivative work, and the project has grounds to claim. Without copyleft, the only possible defense is goodwill.
The side effect also matters. Publishing the derived code exposes its provenance to anyone who compares functions, comments and design decisions. Copyleft does not detect plagiarism, but it makes it expensive to hide, and that is a more honest argument than presenting the license as a filter against AI.
What does anyone looking at the repositories today find?
The project's organization on GitHub offers a snapshot of the change midway through. m1n1, the boot manager used to experiment with Apple Silicon, already appears with a GPL-2.0 license, while the installer data package is still marked as MIT and the video driver appears as GPL-3.0. The Asahi Linux organization groups 43 repositories, and not all of them are moving at the same pace.
That mix fits an ongoing migration without a statement. Changing a repository's license requires editing the license file, the headers of each file and, in many cases, asking permission from contributors who contributed code under MIT, because a license is not revoked retroactively at the maintainer's will.
The project's homepage still has no entry on the matter, and it talks about installing from macOS and about Fedora Asahi Remix. Anyone who wants to know which repositories are included and from what date will have to wait for the announcement or look file by file.
Omarchy and DHH, the name that hovers over it
The summary that made the change known points to Omarchy, the desktop distribution maintained by DHH, as the case that may have triggered the decision. There is no public confirmation that this project has used unpublished Asahi code, so the accusation remains in the realm of suspicion.
What has been said is the reaction. DHH stated in an interview that he does not care about the license decision, a response that fits his usual discourse on free software and leaves the debate where it was. Without parties confirming the episode, the discussion moves between licensing policy and anecdote.
For the rest of the ecosystem, the detail has more reach than the spat. If a reference project changes its license for fear of AI rewriting, others may follow the same path, and that tilts the balance toward copyleft in layers where permissiveness previously ruled.
What changes for Asahi users
For anyone running Fedora Asahi Remix on a Mac, the change alters nothing. They still receive binaries and updates, and GPL obligations fall on whoever distributes, not on whoever installs and uses the system on their machine. The guide to cooking your own kernel is a good example of domestic use that the license does not touch.
The real impact shows up for anyone who reuses the code. A company that had integrated Asahi code into a closed product will have to review which license each file carried before the change, because what was already published under MIT does not become GPL by magic. And anyone who wants to contribute from now on will find a different regime from the one they accepted when they started taking part.
There is also a knock-on effect for the developers who package it: Ubuntu, Debian and Fedora keep their own licensing paperwork, and a change like this forces them to review paths, notices and metadata. It is plumbing work that shows up months later, when an automated license checker starts flagging packages.
Generated code, unresolved authorship
Rewriting with AI models raises a question that licenses only partly answer. Derived code that no one can compare line by line is hard to challenge, and copyleft requires publication before distribution, not a declaration of which tool was used to write it. Traceability still depends on good faith.
Even so, moving an entire project to GPLv2 is the most forceful response a small community has given so far to a practice that was becoming normalized. It does not stop the rewriting, it raises the price of distributing it in closed form and it puts maintainers' work where it belongs: under a license that requires something to be given back.
The arrival of CachyOS on WSL shows how far open source reaches on the desktop today. The debate over kernel mitigations shows the other extreme: open source gains users and loses control over what happens afterwards. Choosing a license is the only lever a project has to influence that afterwards.