A key challenge in the study of mutualism is to understand the mechanisms that prevent cheating. In some systems, host retaliation against cheaters prevents the breakdown of cooperation. Here, we focus on the converse of this demonstration, and ask whether hosts that fail to retaliate are commonly inhabited by cheaters. We do so using the classic ant–plant interaction, in which plants provide ant-housing (domatia) in return for protection from herbivores. Our model system is the rattan ant-palm Korthalsia furtadoana, which grows swollen leaf sheaths as domatia and associates with two species of obligate host-ants, Camponotus sp90 and C. sp93, and with facultative Crematogaster and ‘tramp’ ant species. One ant-tree species is known to retaliate by tying the growth of domatia to the successful protection of new leaves, and non-protecting cheaters are rare. In contrast, K. furtadoana grows the domatium before the new leaf develops, suggesting that sanctioning may not be possible. We experimentally simulated herbivory by cutting leaves from shoots and found no difference in the mortality and growth of domatia on such ‘cheated’ shoots than on controls, confirming that K. furtadoana cannot sanction non-protectors. We then investigated the intensity of protection that Camponotus and Crematogaster ant-symbionts provide K. furtadoana. We demonstrate that C. sp90, which only inhabit half of colonised plants, vigorously protects leaves, that C. sp93 rarely protects, and that Crematogaster never protects. We then show that plants inhabited by C. sp90 have a higher growth rate than those inhabited by C. sp93. We conclude that C. sp90 is a protection mutualist, while C. sp93 and Crematogaster are parasites, the first such demonstrations for an ant–palm interaction. The presence of commonly occurring parasites, as well as rare tramp ants, provides the first clear correlative evidence that an inability to punish results in abundant cheating.