This paper discusses the ethical aspects of secondary use of language data: data that have been created and stored for purposes such as administration, training or quality monitoring. Such recordings provide interesting natural language data that stay clear of the observer’s paradox (Labov, 1972). Additionally, they offer opportunities for the field of applied linguistics that seeks to understand everyday language use and to assist language users. However, obtaining informed consent from interactants can be extremely difficult if gatekeepers shield them from researchers for reasons of professional ethics or because of legislation. This article draws on conversation analytic research into police interviews with child witnesses to argue a) that conceptualizing pre-recorded data as archival data is more appropriate than applying the ‘human participants’ ethical model; and b) that if sufficient protective measures are taken, secondary use of data without informed consent can be ethically sound.