I expect that with AI, we'll be less able to rely on the heuristic of bad grammar to easily detect phishing. That one flaw gave the phishers away so often, and made it so obvious ...
* Bayesian poisoning https://en.wikipedia.org/wiki/Bayesian_poisoning
* Weeding out poor mark candidates https://josephsteinberg.com/why-scammers-make-spelling-and-g...
Easy. As you say all you need to do is to compare and as I occasionally have a glimpse of the spam for some non-English language with a non-Latin script - it's the same.
I noticed how well it understands general parlance after it created marketing style copy for me and I told it to sound "less wanky" and it made it much more to the point.