I have read some oblivious transfer protocols, and I am confused about why they do not mention the malicious behavior that the corrupted sender may encrypt the wrong message $x_0'$ and $x_1'$ so that the honest receiver will receive the wrong result $x_\sigma'$. I think this totally breaks the correctness of OT and I don't know why the malicious secure OT protocol did not mention it. I know that the committed OT can prevent this behavior, but is this not a malicious behavior that one malicous-secure OT should consider?
In MPC we judge security by comparing to the "ideal world" where a trusted third party does the entire computation. In the ideal world, an adversary can send anything it wants as input to the trusted third party. In the ideal world, when it comes to malicious adversaries, there really is no valid sense of a "correct/incorrect input" for the adversary. (For semi-honest adversaries, their input is supplied by the external environment and they are bound to use it in the ideal world.) Since "using whatever input you like" is possible on the ideal world, it is not considered an attack against the protocol.
When we are analyzing the security of MPC protocols, we have to be aware that the judgment of some behavior as malicious depends on the enviroment. Another aspect is to consider if there a defense against this (supposably) malicious behavior.
Let me explain: if the receiver of an OT Bob refuses to open it, is he acting maliciusly? That is, if he decided to abort the protocol, is this absolutely an attack? Anyway, is there a defense against abort attacks? How could we force a receiver to open the OT message? In some sense, when a sender Alice encrypted wrong messages, she is aborting the protocol. Analyzing the protocol in its context, if abortion is an attack, you have a problem. Otherwise, this must not be considered in the adversarial model.