Effective management of health crises requires a good diagnosis of the current situation and an update of relevant data. Therefore, reliability of this data is fundamental to implement the necessary measures to manage health crises. In this paper, we study the compliance of COVID-19 cases and deaths reported in Tunisia with Benford’s law. Using a panoply of goodness-of-fit tests, we found that this compliance is disproved. This may raise doubts about the manipulation of this data. A search for outliers in these data followed this analysis using two machine learning algorithms, namely the local outlier factor (LOF) and isolation forests. The number of outliers in the data has heightened doubts about their reliability.