Adapting ethical frameworks such as value sensitive design (VSD) and ethics by design (EbD) to the specificity of AI systems (Umbrello and van de Poel, 2021; Brey and Dainow, 2023) can be seen as a recent attempt to systematically respond to the more general ideas of AI for Social Good (Floridi et al., 2020) or AI alignment (Dung, 2023). Despite the differences among these frameworks, the motivation stems from the same challenge—to ensure that AI systems promote, bring about or perform desirable ethical values through their own design. I thus propose that this shift to embedding ethics in the design, development and social acceptance of technologies is but evincing a relationship between artefacts and “performative utterances” (Austin, 1962). My hypothesis is that this is because technologies might not be seen anymore as an instrumental means contributing to the fulfilment of a social promise of justice or social good, but rather as the very fulfilment of the promise. From the perspective of VSD and EbD, technologies (should) do justice through their own functioning, in the same way that a performative does what it says through its very saying. However, the question of the performative is not only entirely neglected in the AI Ethics literature, but also in their theoretical precedents (Verbeek, 2005, 2011; Winner, 1980, Latour, 1988; Akrich, 1997) with the exception of Joerges (1999), who uses Austin to object to Winner’s essentialism. This exception justifies my goal in this paper: to show what AI ethics could gain provided that attention was given to Austin’s (1962) and Derrida’s (1972, 1977) reflections of the performative and its ethical character when it takes the form of a promise. To do so, the paper focuses on Austin’s notion of “infelicity,” which aims to signify the parasitic cases in which the performative fails to realize, and on Derrida’s critique that this infelicity is, as a structural possibility, both the condition of possibility and impossibility of every performative. This helps us acknowledge that even if we must try that our technologies promote as many desirable values as possible, the possibility of not doing so is structural. Not acknowledging this structural possibility is a serious problem because it would risk the situation that a technology is effectively violent while being called and deemed responsible. By way of conclusion, I sketch out new avenues for how to think from this ambiguity.