This article examines three key aetiological theories of autism (meta-representations, executive dysfunction and weak central coherence), which emerged within cognitive psychology in the latter half of the 1980s. Drawing upon Foucault’s notion of ‘forms of possible knowledge’, and in particular his concept of savoir or depth knowledge, two key claims are made. First, it is argued that a particular production of autism became available to questions of truth and falsity following a radical reconstruction of ‘the social’ in which human sociality was taken both to exclusively concern interpersonal interaction and to be continuous with non-social cognition. Second, it is suggested that this reconstruction of the social has affected the contemporary cultural experience of autism, shifting attention towards previously unacknowledged cognitive aspects of the condition. The article concludes by situating these claims in relation to other historical accounts of the emergence of autism and ongoing debates surrounding changing articulations of social action in the psy disciplines.