“Online harm” is an umbrella term used to refer to any online activities which may cause harm to individuals or society. Psychological research is integral to the study of individual harms, such as misinformation, cyberbullying or radicalisation, however psychological theory is notably lacking from broader conceptualizations of online harm. Following the PRISMA guidelines, we systematically reviewed 46 taxonomies and typologies of online harm and risk to assess how harm is currently conceptualised. We find that there are four main challenges associated with how harm is currently classified; (i) there are notable semantic conflations between causes of harm and resultant harmful outcomes, (ii) current conceptualisations of harm are unable to anticipate future harms, (iii) contextual features are integral for understanding when harm occurs and, (iv) online risks are often cascading and lead to non-linear harmful outcomes. Based on these four limitations, we propose a novel conceptualisation of online harm which integrates psychological theory with a systems-safety approach in order to understand the precise conditions under which vulnerability to harm occurs. We discuss the value of psychological theory for understanding the mechanisms that lead to vulnerability, and we propose that a psychological approach is best able to anticipate and understand potential future harms. Finally, we provide a case study of how our psychological model can be used to organise and understand the current literature.