The ownership will want humans to punish, humans to trust, humans to manipulate, and humans that can be fired. They'll want a human physically present, representing them as oversight over the AI (a job the capital doesn't want to be doing 24/7). There are serious legal matters involved in high functioning AI systems (and that will dramatically increase over the coming decade), the capital will be extremely worried about liability issues. Human scapegoats are ideal. For now capital will want human style consequence-feedback systems for a tier of human oversight over the AI. It'll probably take a full generation before the capital controlling humans forget / give up on the old systems fully.

It's also worth noting that it's far easier at present to press a human into doing something highly illegal than it is to get Claude or GPT to do so.