Today this is buildable, many models will happily help you glue everything you need together to make swapping in a new version of YOLO to track humans viable.
AI researches are out there worrying about the paper clip problem, about the singularity, about cyber security, about bio weapons, and drug manufacturing.
None of them are thinking about forward looking threat actor models.
I don't know what "forward looking threat actor models" means, but there are entire companies built around AI killing machines. Everybody is not only thinking about it, but doing it.