Hot Take: If your child has autism, unrestricted access to the internet, and a chatbot telling him to kill his parents, it's not Character AI who has failed, it is the child's parents.
It's not the first time it goes horribly wrong; and not having the AI do it doesn't mean people won't just be even worse, see Chris-chan
Him being at all on the spectrum means a bit less than the actual degree at which he is, I may be on it as well (Albeit slightly), had unrestricted internet access, and while I'm not confident on saying I would do it differently than what my parents did, I have many arguments for limiting some of it even for an aware and allistic kid
It's a large scale problem, I can't expect the parents to know the intricacies of everything on the internet, but I'd expect them to try and be aware of those parts their kid does take interest to a degree their competency allows
Regardless, while I've always considered this kind of chatbot something a child should be strictly kept away from, because you can't know when it will tell a child to do this, or to glue pizza for a less dramatic example
I'm okay with it not being marketable to children, but the steps neccessary to limit them further than that make me uneasy