Total Reciprocity Public License

Yeah, the law can be frustratingly subtle. Having spoken to one of the people who helped draft the AGPL, they pushed things about as far as they thought they could at the time. GPL/AGPL-style "copyleft" is a fairly subtle legal hack designed to run on top of pure copyright law, without needing to involve contract law. Apparently they even brainstormed stretching public performance! But it couldn't be plausibly stretched to cover software. And they were careful not to do anything too wildly outlandish that strayed from core copyright law.

As for AI, US courts have ruled that LLM training likely falls under fair use, though with a very important footnote that it might be possible to show "dilution" under the fourth factor. However, the court suggests that any future plaintiffs will need to do their homework carefully:

The court disagreed that market harm could be inferred, as that requires too great an inferential leap: “inferring that Llama (and not just any LLM) can be used to create such books, that it will be used to create such books, that consumers will purchase those books instead of books written by human authors, that consumers will buy those books instead of the plaintiffs’ books in particular, and that Llama is meaningfully better at creating those books because it was trained on copyrighted material.”

I think we're getting to the point where someone might hypothetically win that case, if they actually provided strong enough evidence for each of the steps above. Though it's just as likely that the courts would throw up their hands and refer the whole question to Congress.

If you want to try to prevent LLM training in the US, you may need to go beyond copyright law and consider more traditional commercial software licensing approaches. But one major complication here is apparently copyright preemption, which can apparently drag some contract-based claims back into copyright law. Again, consult a real lawyer.

In the EU, there are some attempts to use Article 4 of the Digital Single Market to provide potentially binding, machine-readable opt-outs:

• TDM·AI
• W3C TDMRep
• Creative Commons Signals (not EU-specific, but the same idea)

There's some debate on whether these approaches would actually be binding in practice, or whether further legislation would be required.

You are probably right as I am not a lawyer. But my attempt is to stop AI leeching work and enforcing copyleft or reciprocity

Even with clear text a small open source developer has no chance of winning against Amazon. And even if they do, it would require so much of their time to effectively stop all development work on a project.

I appreciate the clear warning:

This software is NOT licensed under MIT or Apache. TRPL v1.0 is an experimental, aggressive copyleft license. It treats API interactions and Functional Output (like generated code or AI weights) as derivative works. Proprietary use in microservices or SaaS stacks will trigger a requirement to open-source your entire stack.

but I think it's inadequate. Taking it at face value, it does not trigger a requirement to "open-source" your entire stack. They even say that this is a more stringent requirement than "open source" or "free software." It triggers a requirement that "you must release the Source Code for the entire Combined Work under the terms of this License" which is a more than simply "open-source" across your entire stack. You have to TRPL your entire stack, which is likely not possible for many stacks; if you're using GPL software in your stack, that's incompatible.

It treats API interactions and Functional Output (like generated code or AI weights) as derivative works.

What a derivative work is is defined by copyright law (and how the law has been interpreted by courts), not by the license. You can't extend that willy-nilly.

However, if you Deploy the Program (internally or publicly), you must make the Corresponding Source Code available to:

1. All users interacting with the Program remotely.
2. All individuals within your organization who use the Program.
3. The general public, via a standard network repository.

If I'm reading this correctly, if I run this program in prod, even without modifying it, now I'm on the hook for also hosting a copy of the source code.

With respect to AI, the position of the companies training on corpora of MIT and GPL licensed text is that this constitutes fair use; if it wasn't, they would also be in violation of the MIT and GPL licenses. So the AI clause doesn't really do much.

If you Deploy the Program on hardware (consumer electronics, IoT devices, or cloud infrastructure), you must provide "Installation Information." This includes all authorization keys, cryptographic signatures, or passwords required to install and execute modified versions of the Program on that specific hardware.

I think this means that if I deploy a program on my server I have to give everyone root access to that server.