Post #3043075

2026-05-18 13:13 UTC

@muvlon@hachyderm.io @allpurposemat@mastodon.gamedev.place The US copyright office has said that LLM code cannot be copyrighted because they aren't produced by a human - it has nothing to do with attribution: https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf How is the jury still out exactly?

Replies (1)

@muvlon@hachyderm.io 2026-05-18 16:24

@yoasif@mastodon.social @allpurposemat@mastodon.gamedev.place I was talking about the SC judgment that was widely misreported as ruling LLM output "public domain", I actually wasn't aware of this document by the copyright office, thanks for adding it! This does say that what the AI produces is not itself copyrightable, but that's still different from saying that any output is in the public domain. In particular, they also state that the resulting work is still subject to any copyright of works that were in the training set and that are "recognizable" in the output (probably super tricky to prove in practice, but still). So also according to this, LLMs are not legally able to copyright launder anything you want into the public domain, and any projects that want to incorporate LLM-generated code need to be aware of the legal risks.

Open ##3043076