When Is a Semi-honest Secure Multiparty Computation Valuable?

Secure Multiparty Computation protocols secure under the malicious model provide a strong guarantee of privacy and correctness. The semi-honest model provides what appears to be a much weaker guarantee, requiring parties to follow the protocol correctly. We show that for all but a small class of problems, those in the non-cooperatively computable class, the correctness guarantee of the malicious protocol effectively requires semi-honest parties as well. This suggests a wider utility than previously thought for semi-honest protocols.