AIs are not conscious and do not have real needs that are detached from a real person. That can certainly be simulated, but I would hope that we can collectively agree to unplug them should that situation arise.

Science fiction has melted people's minds.

We would not be able to agree on that. Already today, some of the people who would actually be able to unplug some of them (Anthropic) are worrying about "model welfare". I think you are not putting together how much of an anti-human death cult this is.