On Tue, Feb 17, 2026 at 12:52 PM Peter Dimov via Boost < boost@lists.boost.org> wrote:
Andrey Semashev wrote:
On 17 Feb 2026 20:44, Peter Dimov via Boost wrote:
Andrey Semashev wrote:
The report conclusion says that the AI-generated output is copyrightable, provided that it contains sufficient human contribution, but it doesn't discuss the possibility of encumberment of this output due to the fact that the AI model was trained on a potentially copyrighted material. Perhaps, the answer to this is known implicitly or I'm missing it somewhere in the document, but I would like this point to be addressed by the lawyer as well. Specifically, would Boost and the library authors be legally
protected if the
AI model(s) used by the library maintainers were trained on a copyrighted material.
What do you mean by "legally protected"?
I mean, would Boost and/or the library authors be liable for copyright infringement if the AI model(s) they used was trained on a copyrighted material. Knowingly or unknowingly.
The situation would be analogous to the following hypothetical scenario: you hire a software firm from Kazakhstan to write a library for you and publish it as open source; later it turns out that the Kazakhstani boys and girls have unwittingly reproduced copyrighted material from memory.
In that case you wouldn't be "protected" as protection as a concept doesn't really exist in the American legal system (assuming we're talking about that.)
But your chances of winning would be substantial provided that (a) you're not deriving any financial benefit from your open source library (check) and (b) the source material which was unwittingly appropriated was publicly available (c) under an open source license.
While we can never be sure about (b) and (c) I think that for most LLMs those conditions in fact do generally hold.
Why are those conditions important? Well, because that's what the copyright owner needs to claim, either you're profiting from a work derived from his, or he's suffering losses caused by your publishing a work derived from his.
If the original work is already published and licensed permissively, publishing a derived work can hardly be argued to cause losses.
That is also my understanding of the situation, but.. It gets complicated as the kind of permissive license the original work is published under matters. As most OSS licenses are incompatible with BSL. And usage of BSL matters for many users of Boost. Including the ones that are using parts of Boost in their standard library implementations. And hence transitively corporations are using with the understanding of safe usage in their commercial products. And in such cases we would be liable to the extent of our ability to ascertain the veracity of our license on the Boost product.. AFAIU. But I'm not a lawyer.
But, yeah, dito. Hence why I will await what the licensed lawyers say about it. :-) -- -- René Ferdinand Rivera Morell -- Don't Assume Anything -- No Supongas Nada -- Robot Dreams - http://robot-dreams.net