20 years? more like 5

TLS doesn’t encrypt the host name of the urls you are visiting and DNS traffic is insanely easy to sniff even if you aren’t using your ISPs service.

just a guess, but in order for an LLM to generate or draw anything it needs source material in the form of training data. For copyrighted characters this would mean OpenAI would be willingly feeding their LLM copyrighted images which would likely open them up to legal action.