During episodes of identity-targeted violence, what leads people to aid and protect the persecuted? Prior research on rescue during the Holocaust and Rwandan genocide has pointed to the importance of social networks, but the mechanisms remain unexplored, and the overall frequency of these behaviors, unknown. After surveying 2500 survivors of the 1992-5 Bosnian conflict---the first nationwide survey on wartime rescue in a post-conflict state---I find that cross-group assistance was both widespread and strongly correlated with respondents having more cross-group ties. Yet surprisingly, the strength of those ties does not appear to have had much impact. Most people, it seems, were willing to help not only close friends but even friends-of-friends and acquaintances. Drawing on 160 new interviews with helpers and recipients, I theorize how networks activate cross-group social capital to channel aid to those in need. To address problems in recall bias and demand effects, I validate these findings with three novel data sources: a long-overlooked 1990 census table that disaggregates intermarriage rates by municipality, a newly-completed database covering over 96% of wartime fatalities, and over 130 oral histories collected by multiple local researchers during and after the war. Together, these sources provide compelling evidence for a model that explains how diverse networks lead people to not only save lives but perform everyday acts of kindness during riots, genocide, state repression, and civil war.