Fully agree. What is envisaged here is a self evolving, autonomous agent verse where frequent human intervention may or may not be possible and where the the covenant (three laws) should act as the last guard rail against any harm intended by a self conscious agent or their descendants.
What is envisaged is a boundary necessarily baked into the agent sand box. Essentially a prescription for autonomous agentic frameworks with safety as priority. The fact that OpenAI et al demonstrates boundaries being broken means it was never implemented. We also have to understand a model without agency is no threat - that is why it is the agent that needs control on action and not the model.
