Not sure if that's also category #2 or a new one, but also: Places where AI is at risk of effectively becoming a drug and being actively harmful for the user: Virtual friends/spouses, delusion-confirming sycophants, etc.