This paper probes the definitional shortcomings of private military companies, private security companies, and other forms of non-state security. Existing research struggles to conceptualise different forms of non-state and privatised security, often grouping actors together when they should really be treated as separate entities. I develop a dataset of non-state security actors (NSSAs), combining and extending existing datasets on private military and security companies. I collect data on the services these actors provide and the countries in which they operate. I use unsupervised learning to develop a typology of these actors in a more procedural and structured manner. Using this typology in harmony with existing datasets, I infer the true number of non-state security actors across time and space, and show existing findings on the effects of competition among NSSAs are really only driven by what are considered to be ”private security companies”. This dataset will allow for more nuanced theory building in the study of outsourcing violence.