My opinion is that the data disagrees with you: 1. https://www.psu.edu/news/research/story/beyond-memorization-text-generators-may-plagiarize-beyond-copy-and-paste 2. https://dl.acm.org/doi/10.1145/3543507.3583199 3. https://www.sciencedirect.com/science/article/pii/S2949719123000213#b7 4. https://www.theatlantic.com/technology/2026/01/ai-memorization-research/685552/ 5. Related high profile incident that is very telling: https://www.pcgamer.com/software/ai/microsoft-uses-plagiarized-ai-slop-flowchart-to-explain-how-github-works-removes-it-after-original-creator-calls-it-out-careless-blatantly-amateuristic-and-lacking-any-ambition-to-put-it-gently/
In the US at least, there’s clear legal precedent that LLM fabrications are not copyrightable.
I see many people doubt this says anything about training data copyright, beyond AI user copyright.
This isn't legal advice, I'm not a lawyer.