We study the problem of reliably searching for resources in untrusted peer-to-peer networks, where a significant portion of the participating network nodes may act maliciously to subvert the search process. We present a new method called Halo for performing redundant searches over a distributed hash table (DHT) structure to achieve high integrity and availability levels without affecting the storage and communication complexities of the underlying DHT. Other schemes for redundant searches have proposed new or modiﬁed DHTs with increased storage requirements at nodes, requiring modiﬁcations at all nodes in the network. In contrast, Halo aims to serve as a middleware component, making "black-box" calls of the underlying primitive search operation to eventually provide a new composite search operation of higher assurance. We apply this concept to the popular and well-studied DHT Chord, and demonstrate the efficiency and security of our approach though analytical modeling and simulation-based analysis. For example, we show that for 12% malicious nodes in a network of 100,000 nodes, a regular Chord search fails more than 60% of the time. In contrast, Halo reduces this failure rate to 1%. We show how our scheme lends itself to a recursive version that can tolerate 22% malicious nodes with the same level of success, while regular Chord searches fail more than 80% of the time.