LLM outputs are not copyrightable. At least that’s the current established legal precedent in the US. The only question is whether the user owns the copyright without significantly transforming the output but that’s not really relevant in those specific situation.

I mean otherwise it’s a very slippery slope, effectively it would give Anthropic the ownership of any code generated by its models..