The primary source, quoted

James Calligeros's progress report for Linux 7.3, published November 7, contains the decision in a section called "Learning from history":

"Some time ago, we were made aware of multiple attempts to exploit our permissive licensing to hijack our work, with a direct view to replacing the project and aggressively monetising Linux on Apple Silicon. In some extreme cases, in-development feature code was being lifted straight out of our personal forks and published without attribution or lazily rewritten with LLM assistance to obfuscate its origin. The MIT licence explicitly allows this, and has no carveouts to protect against bad faith actors.

"As such, we made the difficult decision to begin the process of transitioning all of our repositories to the GNU General Public License version 2 or later. Going forward all Asahi Linux code will be licenced under strong copyleft licences only. Those who do not learn from history are doomed to repeat it."

The drivers of the change, per the same report: "the malicious practices of the 90s and 2000s have reared their ugly heads once again, this time driven in large part by both the insatiable appetite of frontier AI labs for high-quality training data, and militant animosity toward the Free Software community."

The report itself names its precedent: Wine spent its first years under MIT while commercial proprietary forks were "marketed and sold by various bad actors," and relicensed to LGPL in 2002 to stop the fork drain and pull contributions upstream. Asahi's original MIT choice was deliberate — built so "our friends at the BSDs" could carry the drivers — and the report says plainly that this is the cost being paid: the rationale no longer holds.

The concrete boundary, verified directly: m1n1's LICENSE on main is now GPL v2. Versions through 1.9.9 remain MIT; 2.0.0 onward is GPL v2+. The kernel drivers were always GPL because the Linux kernel is. The transition covers the boot chain, installer, and tools — the parts the kernel's license did not already fence.

What the fork looks like

heise's reporting fills in the ecosystem the report declines to name:

  • Aurora Silicon was founded July 31, 2026 and published repositories for installer, kernel, m1n1, U-Boot, and Mesa through August. Its own attribution page names Asahi as the origin of its bootloader, kernel, U-Boot, and installer trees. The m1n1 fork was renamed Aura; Aurora's own contributions are GPL-2.0-only, while inherited Asahi code remains MIT.
  • Aurora is the technical base of Omarchy M, David Heinemeier Hansson's Apple Silicon effort — fifteen people, introduced September 11, "Let's finish what Asahi started" — and Omarchy 4.5 shipped a native macOS installer on November 4 covering M1/M2/M3 and the MacBook Neo. The Omacom Foundation reports more than $20 million in pledges, including from Meta Superintelligence Labs, OpenAI, and SpaceXSI — with several pledgers committing model token quotas rather than cash.
  • Aurora's policies state its AI and decompilation posture differs "significantly" from Asahi's: AI-generated contributions are permitted if disclosed in the commit, Aurora's output is to be treated as unusable by Asahi, and Aurora contributors are barred from submitting bugs, pull requests, or insights to Asahi. Asahi's copyright rules exist for the opposite reason — a clean-room line (no disassembly of Apple's GPU drivers, decompilation restricted to designated contributors) that keeps the code defensible against Apple.

Three days passed between the Omarchy-M installer's native-Mac launch and the license announcement. Nobody has publicly demonstrated that Omarchy or Aurora specifically is the fork that lifted unattributed code from private forks; the timeline is an inference drawn by heise, and this post carries it as exactly that.

Who answers, and how

Hansson's response on X: gratitude for Asahi's Apple Silicon work, the claim that extending it is "totally in the spirit of open source" — Touch ID, the M3 GPU, DisplayPort over USB-C, deep sleep, the MacBook Neo — and, on Asahi's refusal to upstream what agent-accelerated work now produces, a shrug emoji: the license change "changes nothing."

That last part is technically right for one specific case: Aurora publishes its code. A fork that conveys source under GPL is exactly the license working as designed — the fence is against the quiet incorporation of future Asahi work into products nobody can read, not against forks per se, not against money, and not against publishing. The MIT-licensed body of work up to the cutoff is out there forever, in the same way every permissive grant in this site's ledger is permanent.

OpenBSD is the collateral. The BSDs can take GPL code into their base only with great reluctance — OpenBSD avoids it — and OpenBSD has booted on Apple Silicon through the Asahi installer to this day. Future Asahi drivers are now closed to them, and Aurora's GPL'd bootloader does not help. Bryan Steele's objection, as relayed by heise: the move is disingenuous and misguided, breaking bridges "without necessity" and without the consent of all previous contributors — the fork fight being carried out on the back of a third party that did not cause it. That is the real cost of the change, freely accepted, and it deserves to be stated without being argued away.

The proof problem, which is the actual story

Scott Chacon's reaction — "software licenses are dead" — lands where the evidence points. Copyright protects expression, not ideas; whether an LLM-paraphrased version of file A constitutes a derivative work of file A is juridically unresolved; and Asahi, like most community projects, requires no copyright assignment, so each developer owns their code and only that developer can sue, if they can prove provenance in a diff-erased world — a model rewrites foreign code in minutes until no diff shows the origin.

So the license change cannot deliver an enforcement remedy against the exact technique it names. What it can do is three things, and they are not nothing:

  • It states the contract. Any future incorporation of Asahi code into a distribution carries the LGPL/GPL obligations forward visibly; "we didn't know" dies as a defense on contact with the license file.
  • It changes what silence costs. Under MIT, taking code and publishing a rewrite is free. Under GPL, the same act requires disclosing what you received from whom — and the disclosure paper-trail is what makes the next disputed case provable.
  • It prices the fork. A fork can still take each future improvement, but every take now carries source obligations that a monetized product must service. Aurora already does. Competitors that don't, now inherit obligation instead of immunity.

heise's structural summary deserves the last word on feasibility: a funded team with agents writing code around the clock "could probably have gotten far without Asahi's groundwork" — but the drivers, boot tricks, and hardware knowledge it forks are five years of reverse engineering, and no license can make a volunteer project outrun that. The frustration of being passed by a funded fork wearing "finish what Asahi started" is the honest center of the story, and the license is the answer the project gave to being erased, not a weapon against being passed.

The honest column

  • The private-fork theft narrative has no public case files. Asahi's statement covers attempts it was made aware of; heise draws the Omarchy/Aurora timeline as an inference; nobody has shown which fork took what. Report the timeline as an inference.
  • Pre-cutoff code is MIT forever. The transition changes the forward terms only; an LLM can still paraphrase the MIT corpus of yesterday without legal exposure.
  • The BSDs pay for a fight they did not pick. Whether the cost is "unfortunately necessary" or "unacceptable collateral" depends on how much credit you give the theft narrative — this site holds that tension open rather than resolving it with a slogan.
  • Asahi chose MIT for ecosystem reach and is now retracting that gift because it was used as designed, not because anyone broke the old license. Copyleft in the AI era is a provenance defense at best; the attribution chain, not the license text, is where it will be won or lost.
  • Hansson's position and Asahi's position are both internally consistent. Two communities now hold different definitions of "open source in spirit" — gratitude-and-extension versus attribution-and-fence — and the definitions are not going to be merged. The Claude Code settings-census this site published Monday is the same AI-era provenance problem pointed at secrets instead of source: agents write, models transform, and the records of where things came from are the only remaining evidence anyone can act on.

For your deployment

  • Treat upstream license changes as a governance event, not a changelog line. The m1n1 pattern — old versions stay MIT, 2.0.0+ is GPL v2+ — is the shape most transitions will take. Know which of your pinned components carries which terms per release, or an audit will find the answer for you.
  • If your community maintains patches or forks of anything load-bearing, keep provenance that survives an LLM rewrite: commit-attached author identity, merge-base records, and signed upstream references. These are the records a future dispute will actually need, because the diff will not suffice.
  • When you standardize community infrastructure on FOSS, read the maintainers' license as a sustainability signal, not a legal formality. A project that stays permissive while being forked for money is spending credibility it may not have in two years; a copyleft flip like this one usually means the maintainers have decided being unpleasant to extract is worth more than being loved by every ecosystem.
  • If you are the fork: publishing your source — as Aurora does — is the minimum the license asks, and it is also, not coincidentally, the only way to make the attribution claim in your own README credible. The GPL does not make forks good neighbors. It makes them legible.

Sources: Asahi Linux: Progress Report — Linux 7.3 (published 2026-11-07, James Calligeros); m1n1 LICENSE on main (GPL v2, verified November 11, 2026) and m1n1 releases (v1.9.9, 2026-10-02 — last MIT release); Aurora Silicon: attribution and policies; Asahi copyright rules; heise online, Axel Kannenberg: Asahi Linux ändert Lizenz (November 10, 2026), English version; heise on the Omacom Foundation, August 2026; Hansson and Steele and Chacon responses as relayed by heise's verification, primary links inline; this site's Claude Code settings census (November 10, 2026).