Data-driven networking in combination with machine learning is a powerful way to design and manage networked systems. In this paper, we consider the case of participatory collection of wireless traffic, which is an inexpensive way to infer the wireless activity in a locality. Since such a type of measurement system leans on the goodwill of the end users, it opens a new venue for malicious actions. Possible consequences of attacks are changes in the underlying communication substrate or even the collapse of the network. We assess the influence of these adversaries by identifying possible hostile actions and propose a method to detect them based on unsupervised machine learning models. Through an experimental campaign in various scenarios, we show that attacks with critical impacts are systematically detected, while unidentified attacks produce only a negligible impact in the measurement system.