I’ve never felt comfortable using LLMs to generate code for anything that gets released to the public. Not just for provenance and licensing reasons (although those alone should give you pause), but also for copyrightability reasons. AI art is already deemed un-copyrightable in the US (https://www.theverge.com/policy/887678/supreme-court-ai-art-copyright)—not that that stops even large companies from generating AI slop, which boggles my mind—so it seems to me that we are a small step away from finding out that no code generated by LLMs can be copyrighted at all.
The way I (not a lawyer) see it, using LLMs in my (at least published) software projects puts me in a risky legal situation. I simply don’t do it.