The idea is that if it's solely corporate owned maybe AGI happens a few years later than if it was open-sourced, which gives time for alignment research to potentially come up with a solution to creating a safe AGI, which should be harder than just creating an AGI with no safety guaranteed.

I'm somewhat doubtful that useful alignment research can really be done without actually knowing what the exact architecture of the AI is, but I can see how it would be positive on net if you think unsafe AGI is really likely and really dangerous.