This article aims to acknowledge and articulate the notion of "humanitarian experimentation". Whether through innovation or uncertain contexts, managing risk is a core component of the humanitarian initiative – but all risk is not created equal. There is a stark ethical and practical difference between managing risk and introducing it, which is mitigated in other fields through experimentation and regulation. This article identifies and historically contextualizes the concept of humanitarian experimentation, which is increasingly prescient, as a range of humanitarian subfields embark on projects of digitization and privatization. This trend is illustrated here through three contemporary examples of humanitarian innovations (biometrics, data modelling, cargo drones), with references to critical questions about adherence to the humanitarian "do no harm" imperative. This article outlines a broad taxonomy of harms, intended to serve as the starting point for a more comprehensive conversation about humanitarian action and the ethics of experimentation.