Right - I see the other topic has been locked / author hasn’t returned to follow community rules.
Here’s the original topic https://news.ycombinator.com/item?id=49003386
and a measured (IMHO) response to it. (BTW, do your self a favour and change tabs with that site open :)
https://マリウス.com/i-regret-migrating-to-codeberg/
It’s a tough spot, Codeberg has found themselves in and I wish them luck. But beyond that, this is (yet) another reminder that in 2026, if you don’t self host it, the cloud is just someone else computer
Not true, this does not occur frequently. This study and software LiCoEval from 2024 found 0.88% to 2.01% of code “strikingly similar to existing open-source implementations”. Afaik this is mostly textbook examples, snippets from stack overflow snippets or common github repositories, often replicated api examples and language boilerplate. How you prompt and refine also matters, and for generating novel code or business logic the LLM simply cannot use memorized snippets.
Presumably since then LLMs have worked to reduce that number of memorized code. Since LLMs cannot memorize all their training data, that number is limited. LiCoEval can find the often memorized examples and train to remove them, or suppress them, or they find other ways to reduce direct reproduction from memorization. For example it would be possible to do what malus.sh does with all the training data. Then it cannot memorize copyrighted code.
So for a model that came out 2026 this already small number might not be that relevant anymore.
Copyright can also apply to code rewritten in another programming language, if the overall structure and functionality is identical. Although tbh, that is a bit of a grey area, and in previous rulings there is some precedence that if the “translation” involves significant human effort then it can be considered to be sufficiently transformative. But promting an LLM to “rewrite this in Go” very likely doesn’t pass that relatively high bar.
The study you linked is interesting (and likely underestimates the issue*), but it also shows that it is possible to automate detection of copyright infringing code in LLM output, and even just 2% is more than enough to open a case for malicious litigation. Large companies can write that off as the cost of doing business, but for small individual open-source developers this is a significant risk.
‘*’ Because human programmers usually reuse existing licensed code as libraries, while LLMs are prone to rewrite them, and as explained above a rewrite is at an elevated risk of being considered copyright infringing.
You can also set a LLMs to completely rewriting code snippets from scratch to perform the same function, writing and fulfilling the same unit tests. Not just rewrite in a different language or changing variable names. Basically the transformative abilities of LLMs can make copyright obsolete.
That is wishful thinking. Copyright was originally designed to protect human creativity, and a lot of the laws and precedence cases involve human effort specifically to distinguish between plagiats and original works.
As long as courts don’t start to rule otherwise, LLM generated code is not considered to be copyrightable itself, thus it can not be original work as defined under copyright laws. This in turn lowers the threashold for such code to be considered a derivative regardless of if there is a clear pathway from one to the other. All it takes is someone to claim that it is derivative, and the burden of proof that it isn’t largly falls onto the person accused. And if the code is functionally identical and with no significant human involvement, that is very hard to proove.
IANAL of course.
The IP law is purely a function of economic power, so we know how that will go. To the maximum benefit of the plutocrats that rule our global civilization.
But the more interesting part is the philosophical discovery: If a piece of code can be generated from scratch by a machine that is not creative, then any comparable piece of code is also not creative. In other words, only the very first time a new type of code or algorithm is written can there be creativity - if one follows your argument that LLM machines cannot be creative. This essentially raises the bar for copyright, and makes an objective test of what is copyrightable or not possible. If you can describe something in a prompt, and get a result of comparable function and quality to a human written version, that code is not creative either, therefor not copyrightable.
Obviously we’re not quite there yet, but the more LLMs improve, the less code becomes copyrightable.
Coding becomes pure craftsmanship, like cutting and sanding some lumber and nailing it together to build a shed.
So if we were a rational and logical species, the overwhelming amount of copyright would be deleted and become public domain. Because it is now worth only the cheap solar power needed to generate it.
It is not my argument, but the official stance of the US copyright office and some other similar bodies around the world.
But yes, it poses some interesting philosophical questions, but those are not of very high relevance to copyright law.