For the operative who is confused about polite, Talibany behavior, the senior leadership has been kind enough to offer a code of conduct.  It is written in a spartan and logical fashion, point by point, a bit like the work of Ludwig Wittgenstein.

Number 15 is their policy on torture:

If some one admits that he is spy because you forced or tortured him, that does not make this person a spy and you can’t punish him. lt is prohibited for a Mujahid to promise to someone that if he admits then he will not be killed, he will be let go, or will not be tortured. There are two kinds of promises: the first is forcing, like you are telling him if you admit then we will let you go or we are not going torture you or put you in jail. lf you use force to cause admission, this is not legitimate. Second, you do not use force but you tell him that if you admit we will give you money or a high ranking position. This method also is not legitimate.

They recognize the possibility of false confessions using stick-based incentives like torture.  Interestingly, the humble Mujahid is not even allowed to use carrot-based incentives. There is a problem of generating false confessions from an informant who just wants to get a reward.

The Taliban also know a little elementary implementation theory.  One method to determine if someone is a spy is if (point 14): “There are two witnesses
that testify such person is a spy” (my emphasis).  We can improve incentives to tell the truth if we can cross-check what one person reports against what another reports.  If only one person says someone is a spy, there is the possibility the informant is lying for some reason – personal animosity, theft of the purported spy’s possessions once he is killed or incarcerated etc.  But if we require two informants to say the person in question is a spy, if the two contradict each other, the senior Taliban can at the very least investigate further.  The simplest way to encourage truthtelling is to punish both informants if they contradict each other.  The document does not elaborate on this and relies on ambiguity of outcome (fear of repercussions?) to suffice to give incentives.  The contract is incomplete in other words….another issue that is a concern of mechanism design and contract theory.

I didn’t see any repeated games insights though.  Maybe in the next version.