You can promote those values, but all it takes is one person that doesn't operate by the same ethical system as you to invalidate everything. In the case of MAD that means the annihilation of the world. The more people the technology spreads to, the more chances it has to fall into one of those hands. And unfortunately, those people are always going to exist among the ~8 billion humans alive, somewhere.

All this brings me back to the Vulnerable World Hypothesis proposed by Nick Bostrom. In fact global totalitarian surveillance that stamps out all potentially dangerous research is one of his few proposed solutions in his paper. I don't know if I can stomach such a solution, but I think we are living the early stages of such a world that forces such questions to be asked, not one in which we picked out a "black ball" that will make us go extinct, but a sort of "green ball" that irreversibly poisons specific parts of collective humanity.

I would personally be in favor of OpenAI keeping GPT-4 and future models proprietary, much as it won't affect the open source spirit. Random hackers with powerful GPUs generating terabytes of SEO spam are less visible to the world and more difficult to hold publicly accountable than a huge corporation with billions in funding.