# Malicious party security model: When one party is corrupted at a time

Consider we have three parties,namely client $A$, client $B$ and a server. Client $B$ and the server can be malicious, but we do not allow the adversary to simultaneously corrupt the server and client $B$ (similar to [1]). The scenario is that both clients outsource their private data to the server. Then they ask the server to do computation and send back the result and a proof of computation correctness to client B. I have designed a protocol that satisfy the above properties but the malicious server may send an incomplete result. This can be detected by client B but after that if client $B$ proceeds it can learn client $A$'s data (and this is not desirable).

Question: Based on the above security model, is client $B$ allowed to take further action to figure out the other client's data, after it detects the server's malicious behavior?

Edit: TBN: The outsourced data is not leaked to the malicious server.

• I would say that under that security model, if you are assuming the server is being malicious, then client B cannot proceed with any malicious action. In other words, that would be equivalent, in the security proof, to an adversary who is corrupting both the server and client B at the same time, which contradicts your initial assumptions. – cygnusv May 7 '15 at 10:45

$A$ has outsourced it's private data to the server, right? In that case $A$'s private data has already been leaked to the adversary when the server was corrupted. So there is no added leakage in $B$ also learning this data. In fact since only one of $B$ and the server can be corrupted (assuming static corruptions) we can conclude that $B$ is honest, and there is really no problem.
• Even then there is probably no problem. The argument that $B$ must be honest still holds. So $B$ learning $A$'s data should not be a problem, since what we are trying to do is to protect ourselves from the adversary, not the honest parties. – Guut Boy May 7 '15 at 12:12