Humans closely monitor others' cooperative relationships [1, 2]. Children and adults willingly incur costs to reward helpers and punish non-helpers-even as bystanders [3-5]. Already by 3 months, infants favor individuals that they observe helping others [6-8]. This early-emerging prosocial preference may be a derived motivation that accounts for many human forms of cooperation that occur beyond dyadic interactions and are not exhibited by other animals [9, 10]. As the most socially tolerant nonhuman ape [11-17] (but see [18]), bonobos (Pan paniscus) provide a powerful phylogenetic test of whether this trait is derived in humans. Bonobos are more tolerant than chimpanzees, can flexibly obtain food through cooperation, and voluntarily share food in captivity and the wild, even with strangers [11-17] (but see [18]). Their neural architecture exhibits a suite of characteristics associated with greater sensitivity to others [19, 20], and their sociality is hypothesized to have evolved due to selection against male aggression [21-23]. Here we show in four experiments that bonobos discriminated agents based on third-party interactions. However, they did not exhibit the human preference for helpers. Instead, they reliably favored a hinderer that obstructed another agent's goal (experiments 1-3). In a final study (experiment 4), bonobos also chose a dominant individual over a subordinate. Bonobos' interest in hinderers may reflect attraction to dominant individuals [24]. A preference for helpers over hinderers may therefore be derived in humans, supporting the hypothesis that prosocial preferences played a central role in the evolution of human development and cooperation.