Is legal the same as legitimate: AI reimplementation and the erosion of copyleft

The really interesting question to me is if this transcends copyright and unravels the whole concept of intellectual property. Because all of it is premised on an assumption that creativity is "hard". But LLMs are not just writing software, they are rapidly being engineered to operate completely generally as knowledge creation engines: solving math proofs, designing drugs, etc.

So: once it's not "hard" any more, does IP even make sense at all? Why grant monopoly rights to something that required little to no investment in the first place? Even with vestigial IP law - let's say, patents: it just becomes and input parameter that the AI needs to work around the patents like any other constraints.

Someone should put this to the test. Take the recently leaked Minecraft source code and have Copilot build an exact replica in another programming language and then publish it as open source. See if Microsoft believes AI is copyright infringement or not.

I think the missing thing here is that the license violation already happened. Most of the big models trained on data in a manner that violated terms of service. We'll need a court case but I think it's extremely reasonable to consider any model trained on GPL code to be infected with open licensing requirements.

I believe it is a narrow view of the situation. If we take a look into the history, into the reasons for inventing GPL, we'll see that it was an attempt to fight copyrights with copyrights. The very name 'copyleft' is trying to convey the idea.

What AI are eroding is copyright. You can re-implement not just a GPL program, but to reverse engineer and re-implement a closed source program too, people have demonstrated it already, there were stories here on HN about it.

AI is eroding copyright, so there may no longer be a need for the GPL. GNU should stop and rethink its stance, chuck away the GPL as the main tool to fight evil software corporations and embrace LLM as the main weapon.

From the article:

> He fed only the API and the test suite to Claude and asked it to reimplement the library from scratch.

From GPL2:

> The source code for a work means the preferred form of the work for making modifications to it. For an executable work, complete source code means all the source code for all modules it contains, plus any associated interface definition files, plus the scripts used to control compilation and installation of the executable.

Is a project's test suite not considered part of its source code? When I make modifications to a project, its test cases are very much a part of that process.

If the test suite is part of this library's source code, and Claude was fed the test suite or interface definition files, is the output not considered a work based on the library under the terms of LGPL 2.1?

I agree with the thrust of this article, that norms and what we perceive as good or desirable extend considerably beyond the minimum established by law.

But a point that was not made strongly, which highlights this even more, is that this goes in every direction.

If this kind of reimplementation is legal, then I can take any permissive OSS and rebuild it as proprietary. I can take any proprietary software and rebuild it as permissive. I can take any proprietary software and rebuild it as my own proprietary software.

Either the law needs to catch up and prevent this kind of behavior, or we're going to enter an effectively post-copyright world with respect to software. Which ISN'T GOOD, because that will disincentivize any sort of open license at all, and companies will start protecting/obfuscating their APIs like trade secrets.

> Blanchard's account is that he never looked at the existing source code directly. He fed only the API and the test suite to Claude and asked it to reimplement the library from scratch

This feels sort of like saying "I just blindly threw paint at that canvas on the wall and it came out in the shape of Mickey Mouse, and so it can't be copyright infringement because it was created without the use of my knowledge of Micky Mouse"

Blanchard is, of course, familiar with the source code, he's been its maintainer for years. The premise is that he prompted Claude to reimplement it, without using his own knowledge of it to direct or steer.

In the corporate world, we've started using reimplementation as a way to access tooling that security won't authorize.

Sec has a deny by default policy. Eng has a use-more-AI policy. Any code written in-house is accepted by default. You can see where this is going.

We've been using AI to reimplement tooling that security won't approve. The incentives conspired in the worst outcome, yet here we are. If you want a different outcome, you need to create different incentives.

I think the direction we are going, the GPL is going to fade away. I think people will look at this like writing a book and claiming the ideas in the book cannot be copied. This debate is not that different from the ones going on in the music industry. I open sourced my latest software as Apache 2.0 after debating a lot about this. Unless the FSF wins in court in the next <=2-3 years, there is no coming back from this.

If Blanchard is claiming not to have been substantively involved in the creation of the new implementation of chardet (i.e. "Claude did it"), then the new implementation is machine generated, and in the USA cannot be copyright and thus cannot be licensed.

If he is claiming to have been somehow substantively "enough" involved to make the code copyrightable, then his own familiarity with the previous LGPL implementation makes the new one almost certainly a derivative of the original.

Wow, it feels like this argument rewired my brain.

When I first read about the chardet situation, I was conflicted but largely sided on the legal permissibility side of things. Uncomfortably I couldn't really fault the vibers; I guess I'm just liberal at heart.

The argument from the commons has really invoked my belief in the inherent morality of a public good. Something being "impermissible" sounds bad until you realize that otherwise the arrow of public knowledge suddenly points backwards.

Seeing this example play out in real life has had retroactive effects on my previously BSD-aligned brain. Even though the argument itself may have been presented before, I now understand the morals that a GPL license text underpins better.

This is only worth arguing about because software has value. Putting this in context of a world where the cost of writing code is trending to 0, there are two obvious futures:

1. The cost continues to trend to 0, and _all_ software loses value and becomes immediately replaceable. In this world, proprietary, copyleft and permissive licenses do not matter, as I can simply have my AI reimplement whatever I want and not distribute it at all.

2. The coding cost reduction is all some temporary mirage, to be ended soon by drying VC money/rising inference costs, regulatory barriers, etc. In that world we should be reimplementing everything we can as copyleft while the inferencing is good.

Surprised they don't mention Google LLC v. Oracle America, Inc. Seems a bit myopic to condone the general legality while arguing "you can only use it how I like it".

It also doesn't talk about the far more interesting philosophical queston. Does what Blanchard did cover ALL implementations from Claude? What if anyone did exactly what he did, feed it the test cases and say "re-implement from scratch", ostensibly one would expect the results to be largely similar (technically under the right conditions deterministically similar)

could you then fork the project under your own name and a commercial license? when you use an LLM like this, to basically do what anyone else could ask it to do how do you attach any license to it? Is it first come first serve?

If an agent is acting mostly on its own it feels like if you found a copy of Harry Potter in the fictional library of Babel, you didn't write it, just found it amongst the infinite library, but if you found it first could you block everyone else that stumbles on a near-identical copy elsewhere in the library? or does each found copy represent a "Re-implementation" that could be individually copyrighted?

It should be noted that the Rust community is also guilty of something similar. That is, porting old GPL programs, typically written in C, to Rust and relicensing them as MIT.

It seems that this chap didn't go and implement a new library, he reimplemented an existing one and became sole-controller of it. i.e. he seems to have taken its reputation, brand whatever you call it away from the contributors and entirely to himself. Their work of establishing it as a well known solution is no longer recognised.

So of course we feel that something wrong has happened even if it's not easy to put one's finger on it.

There's a Japanese version of that page, written in classical text writing direction, in columns. Which is cool. Makes me wonder, though - how readable is it with so many English loanwords which should be rotated sideways to fit into columns?

"Antirez closes his careful legal analysis as though it settles the matter. Ronacher acknowledges that “there is an obvious moral question here, but that isn't necessarily what I'm interested in.” Both pieces treat legal permissibility as a proxy for social legitimacy. "

This whole article is just complaining that other people didn't have the discussion he wanted.

Ronacher even acknowledged that it's a different discussion, and not one they were trying to have at the moment.

If you want to have it, have it. Don't blast others for not having it for you.

Broadly speaking, the “freedom of users” is often protected by competition from competing alternatives. The GNU command line tools were replacements for system utilities. Linux was was a replacement for other Unix kernels. People chose to install them instead of proprietary alternatives. Was it due to ideology or lower cost or more features? All of the above. Different users have different motivations.

Copyleft could be seen as an attempt to give Free Software an edge in this competition for users, to counter the increased resources that proprietary systems can often draw on. I think success has been mixed. Sure, Linux won on the server. Open source won for libraries downloaded by language-specific package managers. But there’s a long tail of GPL apps that are not really all that appealing, compared to all the proprietary apps available from app stores.

But if reimplementing software is easy, there’s just going to be a lot more competition from both proprietary and open source software. Software that you can download for free that has better features and is more user-friendly is going to have an advantage.

With coding agents, it’s likely that you’ll be able to modify apps to your own needs more easily, too. Perhaps plugin systems and an AI that can write plugins for you will become the norm?

> Blanchard's account is that he never looked at the existing source code directly.

That’s a weird statement while releasing the new version of the same project. Maybe just release it as a new project, chardet-ai v1.0 or whatever.

You can't put a copyright and MIT license on something you generated with AI. It is derived from the work of many unknown, uncredited authors.

Think about it; the license says that copies of the work must be reproduced with the copyright notice and licensing clauses intact. Why would anyone obey that, knowing it came from AI?

Countless instances of such licenses were ignored in the training data.

Well, the license change sounds pretty strange, but to be honest if I were to use this software I would use it without adhering to the MIT. It's machine-created content which is not, in general, copyrightable. You can assert whatever license you want on such content, but I am not going to adhere to it. For example, I declare you may use the following under the Elastic License

    The

> an argument for protecting that test suite and API specification under copyleft terms.

If we protect API under copyright, it makes it easier to prevent interoperability. We obviously do NOT want that. It would give big companies even more power.

Now in the US, the Supreme Court that the output of an LLM is not copyrightable. So even a permissive licence doesn't work for that reimplementation: it should be public domain.

Disclaimer: I am all for copyleft for the code I write, but already without LLMs, one could rewrite a similar project and use the licence they please. LLMs make them faster at that, it's just a fact.

Now I wonder: say I vibe-code a library (so it's public domain in the US), I don't publish that code but I sell it to a customer. Can I prevent them from reselling it? I guess not, since it's public domain?

And as an employee writing code for a company. If I produce public domain code because it is written by an LLM, can I publish it, or can the company prevent me from doing it?

This article is setting up a bit of a moving target. Legal vs legitimate is at least only a single vague question to be defined but then the target changes to “socially legitimate” defined only indirectly by way of example, like aggressive tax avoidance as “antisocial”— and while I tend to agree with that characterization my agreement is predicated on a layering of other principals.

The fundamental problem is that once you take something outside the realm of law and rule of law in its many facets as the legitimizing principal, you have to go a whole lot further to be coherent and consistent.

You can’t just leave things floating in a few ambiguous things you don’t like and feel “off” to you in some way- not if you’re trying to bring some clarity to your own thoughts, much less others. You don’t have to land on a conclusion either. By all means chew over things, but once you try to settle, things fall apart if you haven’t done the harder work of replacing the framework of law with that of another conceptual structure.

You need to at least be asking “to what ends? What purpose is served by the rule?” Otherwise you’re stuck in things where half the time you end up arguing backwards in ways that put purpose serving rules, the maintenance of the rule with justifications ever further afield pulled in when the rule is questioned and edge cases reached. If you’re asking, essentially, “is the spirit of the rule still there?” You’ve got to stop and fill in what that spirit is or you or people that want to control you or have an agenda will sweep in with their own language and fill the void to their own ends.

> If source code can now be generated from a specification, the specification is where the essential intellectual content of a GPL project resides. Blanchard's own claim—that he worked only from the test suite and API without reading the source—is, paradoxically, an argument for protecting that test suite and API specification under copyleft terms.

This is an interesting reversal in itself. If you make the specification protected under copyright, then the whole practice of clean room implementations is invalid.

> When GNU reimplemented the UNIX userspace, the vector ran from proprietary to free. Stallman was using the limits of copyright law to turn proprietary software into free software. […] The vector in the chardet case runs the other way.

That’s just your subjective opinion which many other people would disagree. I bet Armin Ronacher would agree that an MIT licensed library is even freer than an LGPL licensed library. To them, the vector is running from free to freer.

without discussing copyright, I don't believe any of this is copied. Which I think should be the argument that actually matters.

I downloaded both 6.0 and 7.0 and based on only a light comparison of a few key files, nothing would suggest to me that 7.0 was copied from 6.0, especially for a 41x faster implementation. It is a lot more organized and readable in my armature opinion, and the code is about 1/10th the size.

It's clear that we're entering a new era of copyright _expectations_ (whether we get new _legislation_ is different), but for now realise this: the people like me who like copyleft can do this too. We can take software we like, point an agent at it, and tell it to make a new version with the AGPL3.0-or-later badge on the front.

IMHO, the API and Test Suite, particularly the latter, define the contract of the functional definition of the software. It almost doesn't matter what that definition looks like so long as it conforms to the contract.

There was an issue where Google did something similar with the JVM, and ultimately it came down to whether or not Oracle owned the copyright to the header files containing the API. It went all the way to the US supreme court, and they ruled in Google's favour; finding that the API wasn't the implementation, and that the amount of shared code was so minimal as to be irrelevant.

They didn't anticipate that in less than half a decade we'd have technology that could _rapidly_ reimplement software given a strong functional definition and contract enforcing test suite.

Why are people even having problems with sharing their changes to begin with? Just publishing it somewhere does not seem too expensive. The risk of accidentally including stuff that is not supposed to become public? Or are people regularly completely changing codebases and do not want to make the effort freely available, maybe especially to competitors? I would have assumed that the common case is adding a missing feature here, tweaking something there, if you turn the entire thing on its head, why not have your own alternative solution from scratch?

Not a lawyer, but my understanding is: In theory, copyright only protects the creative expression of source code; this is the point of the "clean room" dance, that you're keeping only the functional behavior (not protected by copyright). Patents are, of course, an entirely different can of worms. So using an LLM to strip all of the "creative expression" out of source code but create the same functionality feels like it could be equivalent enough.

I like the article's point of legal vs. legitimate here, though; copyright is actually something of a strange animal to use to protect source code, it was just the most convenient pre-existing framework to shove it in.

i've been following this for a while.. and the trend for copyright (of any form - books code pictures music whatever) being laundered by reinventing the "same" thing in-some-way.. is kind-of clear.

But what happens with the new things? Has the era of software-making (or creating things at large) finished, and from now on everything will be re-(gurgitated|implemented|polished) old stuff?

Or all goes back to proprietary everything.. Babylon-tower style, noone talks to noone?

edit: another view - is open-source from now on only for resume-building? "see-what-i've-built" style

Why does anyone need his new library? They can do what he did and make their own.

I'm glad we can fork things at a point and thumb our noses at those who wish to cash in on other's work.

I feel like the licenses that suffer the most isn't the GPL, but the ones like SSPL. If your code can be re-implemented easily and legally by AWS using an LLM, why risk publishing it?

It does feel like open source is about to change. My hunch is that commercial open source (beyond the consultation model) risks disappearing. Though I'd be happy to be proven wrong.

Buried in here: Mark Pilgrim suddenly reappearing after his sudden disappearance years ago! Has he been up to anything since then?

I don't think this part is correct: "If you distribute modified code, or offer it as a networked service, you must make the source available under the same terms."

That's what something like AGPL does.

1) Legality and morality are obviously different and unrelated concepts. More people should understand that.

2) Copyright was the wrong mechanism to use for code from the start, LLMs just exposed the issue. The thing to protect shouldn't be creativity, it should be human work - any kind of work.

The hard part of programming isn't creativity, it's making correct decisions. It's getting the information you need to make them. Figuring out and understanding the problem you're trying to solve, whether it's a complex mathematical problem or a customer's need. And then evaluating solutions until you find the right one. (One constrains being how much time you can spend on it.)

All that work is incredibly valuable but once the solution exists, it's each easier to copy without replicating or even understanding the thought process which led to it. But that thought process took time and effort.

The person who did the work deserved credit and compensation.

And he deserves it transitively, if his work is used to build other works - proportional to his contribution. The hard part is quantifying it, of course. But a lot of people these days benefit from throwing their hands up and saying we can't quantify it exactly so let's make it finders keepers. That's exploitation.

3) Both LLM training and inference are derivative works by any reasonable meaning of those words. If LLMs are not derivative works of the training data then why is so much training data needed? Why don't they just build AI from scratch? Because they can't. They just claim they found a legal loophole to exploit other people's work without consent.

I am still hoping the legal people take time to understand how LLMs work, how other algorithms, such as synonym replacement or c2rust work, decide that calling it "AI" doesn't magically remove copyright and the huge AI companies will be forced to destroy their existing models and train new ones which respect the licenses.

One of the things that irks me about this whole thing is, if it’s so clean room and distinct, why make the changes to the existing project? Why not make an entirely new library?

The answer to that, I think, is that the authors wanted to squat an existing successful project and gain a platform from it. Hence we have news cycle discussing it.

Nobody cares about a new library using AI, but squash an existing one with this stuff, and you get attention. It’s the reputation, the GitHub stars, whatever

See also "A Declaration of the Independence of Cyberspace" (https://www.eff.org/cyberspace-independence), and what a goofy, naive, misguided disaster that early internet optimism turned into.

No, AI does not mean the end of either copyright or copyleft, it means that the laws need to catch up. And they should, and they will.

A lot of untagged IANAL takes here today.

I'm less concerned about AI eroding copyleft and more exited about AI eroding copy right.

Someone be brave, and do this to ZFS. Poke the Oracle bear!

LPGL is dead, long live the AI rewrites of your barely open source code

Perhaps software patents may play an even bigger role in the future.

Imagine if the author has his way, and when we have AI write software, it becomes legally under the license of some other sufficiently similar piece of software. Which may or may not be proprietary. "I see you have generated a todo app very similar to Todoist. So they now own it." That does not seem like a good path either for open source software or for opening up the benefits of AI generated software.

Easy solution for now:

Add something like this to NEW gpl /bsd/mit licenses:

'you are forbidden from reimplementing it with AI'

or just:

'all clones, reimpletetions with ai etc must still be GPL'

What if someone doesn't declare that it has been reimplemented using an LLM? Isn't it enough to simply declare that you have reimplemented the software without using an LLM? Good luck proving that in court...

One thing is certain, however: copyleft licenses will disappear: If I can't control the redistribution of my code (through a GPL or similar license), I choose to develop it in closed source.

That's a non-sequitur. chardet v7 is GPL-derived work (currently in clear violation of the GPL). If xe wanted it to be a different thing xe should've published as such. Simple as.