Please, consider two honest parties $A$ and $B$ outsourced their private data to a malicious server $S$. So the parties store their data in the server. Then at a later point in time they want to ask the server to run some computation on both parties data and return a private result back to a party.
In order to prove the security of the scheme,in the ideal world, we construct a simulator $SIM_s$ simulating malicious server, $A_s$.
Question: Does $SIM_s$, pick two random datasets, make them private and send to $A_s$?
What is not clear in this case is that whether the parties (or simulator) have input or the server has input. Because the parties have already outsourced their dataset.