I don't think this is the correct interpretation. I think they mean that if you make something without AI and then modify that with AI, that's covered. Likewise, if you start from an AI output and modify it, that's covered.
But the pure output of a generative model cannot be copyrighted, regardless of how complex the prompt is (note that the prompt itself could be copyrighted).
> But the pure output of a generative model cannot be copyrighted, regardless of how complex the prompt is
If that’s how the court interpreted it, then the software industry is hosed, since that’d mean none of the generated code running in production right now is under any sort of copyright or otherwise protection, lol.
I doubt that much software is entirely AI-generated with no human review or testing, it’s probably more like integrating some public domain snippets you found online into your code (which doesn’t invalidate copyright on the rest of it, or the way it’s put together) or having some files auto-generated by a script (like a C header containing a lookup table for a simple mathematical function, the table isn’t copyrightable itself maybe but the software as a whole still is)
If a deterministic machine transformation from a copyrightable prompt results in an uncopyrightable image, what do you think a compiler is doing to source code?
AI is not specifically not deterministic from the enduser's perspective. they throw randomness into it and hence why an exact prompt wont produce the same exact result.
a compiler on the other hand is generally pretty deterministic. The non determinism that we see in output is usually non determinism (such as generated dates) in the code that it consumes.
because they are just translating code (that everyone agrees is copyrightable) in a deterministic manner into another medium.
I'm not saying AI art should or shouldn't be copyrightable. One can argue the inputs into the AI generator are copyrightable, but if the output isn't deterministic translation of the input, its a different argument.
The original argument was that AI works wouldn't be copyrightable because they are deterministic, i.e. are just an algorithmic transformation lacking in creativity.
It sounds like they might be under the impression that having any AI-generated output in the code even if parts are human authored would invalidate the copyright, which isn’t true
>If that’s how the court interpreted it, then the software industry is hosed, since that’d mean none of the generated code running in production right now is under any sort of copyright or otherwise protection, lol.
I'm not sure this is really true, since copyright applies to distribution.
If you have a substantial amount of backend code (as with most SaaS projects) you're never actually distributing the code, and copyright is never at play. Computer generated artifacts are already in this boat and are protected by virtue of being trade secrets not by copyright.
This could maybe be true of shipping javascript to the browser, which presumably is not going to qualify as a trande secret, but I don't think that's where most companies derive value.
The idea that copyright applies solely to distribution is a popular myth, but it has no support in the actual copyright law. The core exclusive rights in copyright are (in the US, 17 USC § 106):
---
(1) to reproduce the copyrighted work in copies or phonorecords;
(2) to prepare derivative works based upon the copyrighted work;
(3) to distribute copies or phonorecords of the copyrighted work to the public by sale or other transfer of ownership, or by rental, lease, or lending;
(4) in the case of literary, musical, dramatic, and choreographic works, pantomimes, and motion pictures and other audiovisual works, to perform the copyrighted work publicly;
(5) in the case of literary, musical, dramatic, and choreographic works, pantomimes, and pictorial, graphic, or sculptural works, including the individual images of a motion picture or other audiovisual work, to display the copyrighted work publicly; and
(6) in the case of sound recordings, to perform the copyrighted work publicly by means of a digital audio transmission.
---
OTOH, distributing copies created in violation of copyright is a good way to cause legally-cognizable harms to the copyright holder that will increase the potential damage award when you are found liable for copyright infringement, and it also makes it much more likely that someone will notice the infringement in the first place. But its not where the law, on its own terms, begins to apply.
Doing any of those without permission (unless it falls into one of the exceptions to copyright protection, like fair use) is a violation of copyright.
The idea of copyright is to prohibit unauthorized use and reproduction, but none of this actually happens with a proprietary software SaaS backend. You don't actually give anybody the code - they connect to the service.
Access to the service is already governed by computer access laws, which don't depend on copyright. And if you never intentionally distributed your code outside of your org, you can call it a trade secret and nobody else has any legitimate right to access it - whether or not it is copyrightable.
There are other things that aren't copyrightable that are trade secrets already. This would be true of any kind of automated data collection for example. You couldn't copyright it but you can call it a trade secret.
And for any of that stuff, if you want to share it and limit distribution, you just have whoever wants access explicitly agree to be bound by contract law.
>The idea of copyright is to prohibit unauthorized use and reproduction, but none of this actually happens with a proprietary software SaaS backend. You don't actually give anybody the code - they connect to the service.
The point isn't that you have to give it to people, but okay?
>Access to the service is already governed by computer access laws, which don't depend on copyright
Yeah, copyright doesn't control everything, and?
>There are other things that aren't copyrightable that are trade secrets already. This would be true of any kind of automated data collection for example. You couldn't copyright it but you can call it a trade secret.
Okay?
>And for any of that stuff, if you want to share it and limit distribution, you just have whoever wants access explicitly agree to be bound by contract law.
Your point being? You're just rambling assumptions about copyright and other things, which don't even track the actual law.
> Your point being? You're just rambling assumptions about copyright and other things, which don't even track the actual law.
I'm replying to the post that claimed:
> If that’s how the court interpreted it, then the software industry is hosed, since that’d mean none of the generated code running in production right now is under any sort of copyright or otherwise protection, lol.
There is in fact "otherwise protection" for the software industry by... not distributing the code. They don't need copyright over the generated code if they vibe code a SaaS backend. Whether there's copyright or not is irrelevant for the business model.
Copyright is the strongest legal protection available. It does not have a state of mind element. Breach of contract is much more complicated and context-dependent.
>There is in fact "otherwise protection" for the software industry by... not distributing the code.
Copyright protects against reverse engineering in some circumstances, for example.
>Whether there's copyright or not is irrelevant for the business model.
Yeah, I'm going to continue to disagree with you as I'm actually a litigator.
> Yeah, I'm going to continue to disagree with you as I'm actually a litigator.
OK, can you explain to me why this is a disaster for a vibe-coded SaaS? Why are computer access and/or contract laws insufficient and why would a vibe-coded backend be a huge risk?
I really don't understand where copyright on the code itself is necessary to protect these business models, and hopefully you can help fill the gaps.
I didn't say it would be a huge risk, I just disagree that any of those features of the law cover what copyright does. They don't. If a trade secret is ever revealed, all protection is lost. Breach of contract is very complex compared to an infringement claim and would have to be negotiated. As a customer, why would I want to indemnify a software supplier? If there's no indemnity, it's not going to get anyone very far. CFAA basically requires that something get hacked so it's not going to cover the vast majority of scenarios...
>I really don't understand where copyright on the code itself is necessary to protect these business models, and hopefully you can help fill the gaps.
Well, did you ever try to understand? It's so exhausting coming to these threads when people are just making assumptions about how the law works without any regards to what actually happens, and then suggesting policy changes in response.
Here's a scenario - disgruntled ex-employee leaks the code. Now it's free for anyone to use because there is nothing you can do to stop anyone from using it because you have no rights in the code since the trade secret is broken. You can sue the employee. They are probably judgment proof, wont have a lot of money anyway, and will still not stop a competitor from spinning up the same exact thing.
Trade secret was your suggestion by the way... So do, you actually know how trade secrets work, or you just making things up??
What constitutes a modification? Here’s a reverse Sorities Paradox situation.
Let’s say I use an AI prompt to generate an image with 24-bit color, and then I manually change the RGB value of a single pixel from (255,255,255) to (254,255,255).
Does that constitute a modification and would then allow the image to be copyrighted? If not, where is the line?
This is how I understood the original decision a while back - that there had to be some additional element of human involvement post-"gen", though to what extent is still a bit unclear to me.
What's the threshold? Can the person just slap an LUT on an SDXL image in Photoshop and call it a day?
But the pure output of a generative model cannot be copyrighted, regardless of how complex the prompt is (note that the prompt itself could be copyrighted).