This paper describes the results of an experiment in which human participants were required to detect degraded robot swarm behaviour and classify it as arising from either faulty or malicious robot activity in an idealised simulation of a multi-agent search and rescue task. The accuracy of participant judgements was influenced by the nature of the degradation, and between-participant differences in the extent to which they interacted with the swarm did not significantly influence their accuracy. It was found that detecting and classifying swarm degradation are challenging tasks that are likely to be strongly sensitive to task setting and will tend to require careful swarm system design and specific operator training.