Only works when there is a genuine risk of retaliation equal or worse than the damage done, i.e.:

• There are multiple rounds or each round is done is such a way that the second person has a real choice, or in other words people can’t just make a one-sided choice, cash their gains and dissapear but instead face consequences immediatelly or on subsequent rounds.
• Both sides have similar power to inflict hurt on the other side.

In fact various experiments from Behavioural Economics similar to these and done with conditions more like I describe show exactly that effect: far fewer behave nicelly and are generous and forgiving when others can’t meaningfully retaliate.

It’s not by chance that most situations in real life of somebody taking advantage of somebody else either involve a context where the abuser can just extract a gain and dissapear, escaping retaliation, or there is a massive power imballance so the victim can’t actually retaliate in any meaningfull way (say, the “boss” vs an employee that can easilly be fired or an elected politician making choices that hurt a minority of electors).