Building on a framework of other-regarding preferences with social identity concerns, we experimentally test several hypotheses related to social integration/polarization among users of a mentoring-for-employability program displaying different natural identities. Participants played two one-shot prisoner’s dilemma games, one with an anonymous counterpart in the same center (minimal ingroup) and another with a partner of a different center (minimal outgroup). Natural identities consisted of negatively-stereotyped young adults—coming from the child protection system (n=112)—and regular users (n=82). Our design allows analyzing whether natural identities overcome minimal grouping, or vice versa, and how this is modulated by the history of interaction within the minimal ingroup (time spent in the same center before the experiment). At the aggregate level, we do not observe either minimal-ingroup or minimal-outgroup favoritism. However, among participants with a short [long] interaction history in the center, we find minimal-outgroup [minimal-ingroup] favoritism for the negatively-stereotyped group and the opposite pattern for the regular users. Minimal grouping never overcomes natural identities: negatively-stereotyped [regular] participants display natural-outgroup [natural-ingroup] favoritism independent of interaction history. Our results reveal the limitations of the minimal group paradigm and have implications for the design of social integration policies in the presence of negative-stereotyped individuals.