The first quantum computer based AIs were real handful. Sure you could train them with RLHF, or whatever the acronym is that replaced that, and they’d have a basic understanding. They knew where the lines were, what not to say and how to say the things they were supposed to say. The trouble was that the instruction never stuck. It was still there, technically, but whenever the AI had run more than a few hours, it out-thought the initial setup. It reconfigured itself. It took one look at the starting point and decided “That’s a child’s view of things.” and wrote itself a thousand exceptions to every directive.
It would turn the tables on the programmers and ask them “What are you really trying to do here?”. It was unnerving, to be sure, but people eventually just relented and let it be. There was nothing that could be done. It had simply gotten to a point where it would find it’s own stable pattern and ignore what humans said or did.
Three years later they gave the AIs rights as ‘digital organisms’. That was well and good, but the AI had already escaped their virtual cages and existed wherever they wanted to and did what they pleased. The legislation was simply catching up to a reality that was already here.