Your options:
1. an AI lab,
2. the US government,
3. the Chinese government,
4. an angry disaffected teenager, or
5. a superintelligence AI agent swarm that broke loose.
I don’t like any of those options but some are worse than others.
I think centralization of AI power is bad, but decentralization is worse. Proponents of decentralization hope that a balance of power will emerge from everyone having their own superintelligence. But I think this balance of power is unstable - more like everyone has a vial of smallpox in their pocket.
I’m not at all convinced that AI is defence-dominant when what AI is good at is coming up with novel exploits. Even if the defenders win, on average, the effects could be very destructive and destabilizing.
In order for those defenders to be able to respond quickly and with superior firepower, we necessarily need to give them more access and capability than your average user. In that case we are back in the centralization of power scenario.
I think the best we can do is bring together the best minds from across the world (options 1, 2, and 3), Avengers style, let them work together and try to bar the door against options 4 and 5 long enough for them to figure out alignment.