LLMs running out of free/original content (are we going to see an increase in public spying/recording since that might be the only source of original conversation content left?)
AI self-poisoning. There's a push to better identify synthetic data- since training models on content they generated leads to model collapse.
Ethical AI -- this topic area didn't interest me before, but when you start thinking about pre-programming solutions to moral dilemmas (especially trolley problems) and different cultural/religious/ethical views... and then the principles of the person programming those solutions --- there are a ton of fascinating and unexplored questions in this area.