Reading the below article makes you shiver, especially if you think about all the buzz those companies are getting and creating to compete with good old antivirus vendors (which don't really work for other reasons).
If you look at how the below attack works, it is obvious that a model that you train once, with a lot of good and bad data with something that you call AI, but really is just a statistical model will be open to adversarial attacks and there are great examples for very creative attacks on AI (here, here, here).
My personal conclusion is that no matter how good or bad your endpoint security is, it will not safe you from harm, bad things will happen and the underlying problem of attackers targeting people with Links, Office files or just PDFs.
My other personal conclusion is that we can not rely on the silver bullet, we need to implement more controls, we need to score where our biggest risk, our most valuable data and our most targeted users are and we need to figure out how to protect those.
“Their crime is not that they coded AI poorly. Their crime is calling what they did AI."