If we observe the specific functioning of generative artificial intelligences, we can say that they are systems that produce speech, concrete realizations of natural language, written or oral, that are emitted through an interface. These fragments of sensitive materiality (whose emission, we must not forget, depends on a digital infrastructure and a binary code that we do not see) constitute statements in which we can recognize the two levels: that of the statement, that which is the object of the discourse; and that of enunciation, presupposed in every statement to the extent that every discourse comes from an I whose addressee is a you. Starting from this point, we propose to explore in this work in what sense it is possible to speak of enunciation in generative AIs, what differences they present with respect to human enunciation, and what are the conditions of production, circulation and interpretation of this new discursive type in the digital age. In this way, we will justify the need to attribute the adjective “machinic” to show that the acts of speech production carried out by these language models are superficially similar to human speech acts, but radically different in their constitution. This generates high-impact effects on various dimensions of human existence, given the exponentially increasing frequency of exchanges with this type of models in more and more areas of our lives (as proposed and promoted by tech companies).