I have a doubt regarding the definition of (static) semi-honest adversaries. In this tutorial I read:

Such an adversary controls one of the parties (statically, and so at the onset of the computation) and follows the protocol specification exactly. However, it may try to learn more information than allowed by looking at the transcript of messages that it received and its internal state.

Does it means that, for example, in the case of a commitment scheme a semi-honest adversary, who controls the party supposed to commit a value, knows the committed value (since it is part of her or his internal state)? However, since the adversary is semi-honest, she or he cannot take advantage of this information to deviate from the protocol in a profitable way.


1 Answer 1


Yes, this is what it means. A semi-honest adversary knows the entire state of each corrupted party, including its input, any random coin tosses, and all incoming messages from all other parties. However, it cannot deviate from the specified protocol instructions. Note that in the multiparty setting, the semi-honest adversary may corrupt multiple parties. In this case, it sees all of the corrupted parties' states together.

  • $\begingroup$ Thanks for the reply. In the more complex scenario of a secure multiparty voting protocol, analogously, once a semi-honest adversary learns the preference of one of the corrupted parties, she or he cannot change "strategically" the preference of another corrupted party (also if any preference is legit). Is this correct? $\endgroup$
    – Lorenzo
    Commented Jun 2, 2020 at 11:36
  • 2
    $\begingroup$ In the semi-honest model, the adversary cannot change the user's input full stop. In reality, this is too weak for many scenarios, and in particular for voting. There is something called "augmented semi-honest" where the input can be changed. However, in general, you would probably want malicious security. $\endgroup$ Commented Jun 2, 2020 at 16:20
  • 1
    $\begingroup$ And, in reality a semi-honest adversary can easily become a malicious one. The Covert adversary model is more realistic. $\endgroup$
    – kelalaka
    Commented Jun 2, 2020 at 20:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.