One thing I find interesting is that we're moving from "Can AI write well enough?" to "Can you prove a human wrote this?"
Even if the model output is excellent, legal and publishing workflows care about provenance, not just quality.
That feels similar to software supply chains. We don't only care that a binary works—we also care where it came from, how it was built, and whether we can audit it.
I wouldn't be surprised if authors eventually need some form of verifiable writing history, just as developers increasingly rely on signed commits and build provenance.