Some thoughts on AI works, copyright, and licenses.
To start with, my rough understanding of AI works and copyright are that AI works are not copyrightable unless one specifically has like unique creative input as part of the process. I don’t know what the threshold would be since generally I’d guess that the unique composition of a prompt is not enough (otherwise AI art would mostly be copyrightable). I also don’t know how well this precedent has been set or in which jurisdictions.
With software licenses, my understanding is that the right to license software is built upon copyright law. Owning the copyright for some work is prerequisite for legally publishing it under some license.
If we take AI art as a precedent, then simply embedding an idea in a prompt is not sufficient to own copyright. This lines up with my understanding of IP law more generally, in that you cannot own ideas, but you own an embodiment: either a creative work, via copyright, or a design, via a patent.
Therefore, most AI written code is not copyrightable, distributing it under a license is potentially some kind of like copyright fraud (I’m not sure exactly what this would be legally), and integrating it into projects might break the terms of some licenses or is at least a liability for the maintainer (since you typically need to own the copyright of code to contribute it).
This can be dealt with by having some kind of dual licensing, but that is a massive PITA and people probably won’t do it.
If this ever comes up in some court battle, a private entity might be able to argue that a license doesn’t apply because the project was tainted, or that the plaintiff doesn’t have the right to actually bring the case in the first place.
It also means that AI agents are regularly committing some kind of unlawful action by licensing code they produce. I’ve had this happen plenty if using a coding agent to initialize a new project, it’ll just decide that it should be MIT licensed.