I largely view markets as pooled public bounty for information. Not only does it help in delivering the elicitation reward to the person with relevant information, but it also pools resources to pay that person. I don't think AI superforecasters will know exactly who to reach out to for information.
The lack of improved rationality, purely from a betting and capabilities perspective, would be surprising to me in future AI systems. But broadly agree it's a large brush stroke assumption.
It's not clear to me where in your link the formulation of general-purpose prediction markets can be attributed to anyone else. Particular use cases for betting on papacy or a political election isn't necessarily obvious to me.
"I focus on a scenario where an international agency enforces drastic limits to AI development for two years, starting at the beginning of 2028"
What is your p value for this occurring?