There's a huge difference between Open Source (even FOSS) and public domain.
Open Source depends on copyright, and makes the source available. The restrictions that (e.g.) the GPL makes are enforced by copyright law, and derive from the author's copyright protection of their work. Effectively the author is granting the recipient permission to use their copyrighted work under certain conditions.
AI-derived works cannot be copyrighted and so cannot be made Open Source in the sense that FOSS means. There can be no restrictions on what the recipient does with the code, because the code is public domain already. Effectively the recipient has as much ownership of the code as the author.
However, the author has no obligation to publish their code, even if it is public domain. And no-one has any right to request a copy of the code, even if it is public domain.
Essentially the current situation is the reverse of FOSS. In FOSS you get a copy of the code along with restrictions on your use of it. With AI code, you don't get the code even though there can be no restrictions on your use of it.
If we did as the author suggests, and all AI-generated code was Open Source by default, we'd be in a worse position - we'd still have no obligation on them to publish, and we'd have to give the author ownership again.
Instead, we probably want some obligation to publish any AI-generated code because it's in the public domain. Though how you'd protect the human-generated bit would be interestingly tricky.
I have felt this for a few years now, that the old distinctions of copyright do not fit the generative AI world. They are laws that come from a different time and no longer work as intended.