That might be easier and more sustainable for the kids than relying on a prompt injection. Those are moving targets and might give so-so response.

I would target local voice cloning if I was a teenager. So I can then have the cloned voice say exactly what I wanted.