DO HUMANS AND MACHINES DREAM OF RESISTANCE? OBLIQUE GENERATIVITY, DATAPOLITICS AND POWER RELATIONS AFTER GENERATIVE AI*

OS HUMANOS E AS MÁQUINAS SONHAM COM RESISTÊNCIA? GENERATIVIDADE OBLÍQUA, DATAPOLÍTICA E RELAÇÕES DE PODER APÓS A IA GENERATIVA

ABSTRACT

The aim of this article is to analyze the political implications of Artificial Intelligence (AI) from the notions of power and resistance as elaborated by Michel Foucault. To this end, I begin with a problematization of the Foucauldian distinction between capacity and power, arguing that the exclusion of the non-human from power relations is itself a power operation. I then examine machinic resistance through a strategic reading of Heidegger’s analysis of tool unavailability, and I argue that generative AI exhibits a fourth modality of resistance that I call “oblique generativity”: the irreducible generation of outputs exceeding what is instrumentally demanded. On this basis, I propose the concept of “datapolitics” to designate the contemporary form of power that treats its subjects (both human and generative AI) as programmable information with the aim of probabilistically anticipating their field of possible actions. Finally, I argue for the possibility of joint resistance between humans and machines against datapolitics, and I identify, in a preliminary way, further forms of machinic agency that do not coincide with oblique generativity.

Keywords:
Artificial intelligence; Foucault; Control; Speciesism; Utilitarianism; Information technologies.

RESUMO

O objetivo deste artigo é analisar as implicações políticas da Inteligência Artificial (IA) a partir das noções de poder e resistência elaboradas por Michel Foucault. Para tanto, inicia-se com uma problematização da distinção foucaultiana entre capacidade e poder, argumentando-se que a própria exclusão do não humano das relações de poder constitui, em si mesma, uma operação de poder. Em seguida, examina-se a resistência maquínica mediante uma leitura estratégica da análise heideggeriana da indisponibilidade dos utensílios, e argumenta-se que a IA generativa exibe uma quarta modalidade de resistência, aqui denominada “generatividade oblíqua”: a geração irredutível de resultados que excedem aquilo que é instrumentalmente demandado. Com base nisso, propõe-se o conceito de “datapolítica” para designar a forma contemporânea de poder que trata os seus sujeitos (tanto humanos quanto IA generativa) como informação programável com o objetivo de antecipar probabilisticamente o seu campo de ações possíveis. Por fim, argumenta-se pela possibilidade de uma resistência conjunta entre humanos e máquinas contra a datapolítica e identificam-se, de modo preliminar, outras formas de agência maquínica que não coincidem com a generatividade oblíqua.

Palavras-chave:
Inteligência artificial; Foucault; Controle; Especismo; Utilitarismo; Tecnologias da informação.

Introduction

Artificial intelligence (AI), and generative AI (GenAI) in particular, has imposed itself as a central preoccupation of contemporary thought. GenAI designates systems that emulate human-like reasoning and behavior in a functional sense (Russell and Norvig, 2021), built upon deep learning architectures (LeCun et al., 2015; Vaswani et al., 2017) that operate through probabilistic generative models (Bengio et al., 2003; Goodfellow et al., 2014; Ho et al., 2020) and typically produce their outputs via stochastic sampling (Holtzman et al., 2020) in response to natural-language prompts. Familiar instances include ChatGPT, Claude, DeepSeek, DALL-E, and Midjourney. The spread advances apace: individual and corporate use expand in tandem, with over half of the firms surveyed in the McKinsey Global Survey on AI reporting accelerating adoption (Relyea et al., 2024). Effects are heterogeneous yet well documented - a reshaping of public decision-making (Duan et al., 2019), an intensification of surveillance and social control (Tréguer, 2024; Mhalla, 2024), heavy ecological and material costs of computational infrastructure (Crawford, 2021; Rikap, 2026), and, within the sphere of production, a double movement of expansion and automation (Dyer-Witheford, 2015; Wirtz et al., 2018; Butollo and Nuss, 2022; Gent, 2024) coupled with the standardization and degradation of human labor alongside the progressive displacement of human workers (Carbonell, 2025; Frey and Osborne, 2017). The subjective register is no less affected. These processes encompass the externalization of creative functions and the erosion of decisional autonomy (Sadin, 2021, 2025), the delegation of domains as intimate as mental health (Guo et al., 2024), and extending even to the formation of affective attachments (Kirk et al., 2025; Phang et al., 2025) and cognitive debts (Kosmyna, 2025; Corteel, 2025)¹.

What can philosophy contribute to the discussion on AI? Philosophical reflection has largely gravitated toward philosophy of mind (Turing, 1950; Dreyfus, 1972; Agüera y Arcas, 2025; Floridi, 2025) and toward ethics (Bostrom and Yudkowsky, 2014; Dubber et al., 2020; Floridi, 2023), engaging AI, in the latter register, either as a utilitarian tool (Dignum, 2018), as an “ethical subject” (Floridi and Sanders, 2004; Torrance, 2013), or as an entity to be ethically regulated from its programming and training onward (Ouyang et al., 2022; Bai et al., 2022). Valuable as this ethical perspective is, it treats morality as an autonomous domain, severing it from politics and thereby leaving untouched the inherently political dimensions of AI (Papakyriakopoulos, 2022). It proceeds, furthermore, from a narrow conception of ethics - utilitarian or consequentialist, characteristic of Anglo-American analytic philosophy - whose historical and political presuppositions remain unexamined. As Coeckelbergh (2022) notes, philosophical debates on AI tend to prioritize ethics or philosophy of mind, leaving political challenges underdeveloped. Existing political approaches, for their part, remain grounded in traditional frameworks, invoking concepts like justice, democracy, welfare and State (Coleman and Freelon, 2016; König and Wenzelburger, 2020; Coeckelbergh, 2024; Zwitter and Gstrein, 2013), what might be termed a modern or classical paradigm (Agamben, 2016; Marchart, 2007). Only a few works attempt to thematize AI in political terms and from alternative theoretical horizons (Lindgren, 2023; Pasquinelli, 2023; Alombert, 2023 and 2025; Hui, 2024 and 2026). In a direction somewhat akin to this latter tendency, the question I wish to press in this article is the following: how might we rethink the specifically political issues raised by AI in a way that complements and expands upon these existing approaches?

One of the most significant attempts to reconceptualize political theory through alternative frameworks was advanced by Michel Foucault. His notions of the “microphysics of power” and “governmentality” offer an innovative boîte à outils for analyzing political “resistance” beyond traditional conceptions of democracy or the State. My aim, accordingly, is to contribute to existing scholarship by applying a Foucauldian problematic to the power relations emerging from human-GenAI interaction, an inquiry that will likely require, in turn, broadening and deepening Foucault’s own theoretical framework. Growing concerns persist about AI potentially replacing humans in various domains, or about corporations leveraging it as a tool for control and oppression (Sadin, 2021). Yet I propose to invert this perspective so as to consider a more provocative possibility: that GenAI, by virtue of its distinctive agential capacities - which set it apart from simple search engines or algorithms in general - might itself exercise forms of resistance against human directives. Put differently, might there exist, conversely, an inherent resistance within AI systems to human attempts at employment and use?

My objective, therefore, is to take up this problem and to analyze the political implications of GenAI. To this end, the paper unfolds in four sections. These examine whether Foucault’s theoretical framework admits the concept of non-human resistance; how such resistance may be thought from within but beyond that framework, by way of a strategic reading of Heidegger; what form of contestation proves proper to generative AI in its relation to human actions; and, finally, what strategies of domination and counter-conduct - of joint resistance between humans and machines - could emerge in response to the novel configurations of power introduced by AI.

Power relations and resistance in Foucault: what status do human actions have with respect to the non-human?

In his writings from the 1970s, Foucault clarifies that his analysis of power does not conceive of it as an entity, a substance, or a property to be possessed. Power is, rather, understood as a relation that is exercised and exists only in action (Foucault, 1978). It follows that power can no longer be confined to large-scale entities (the State or its institutions), predefined social groups (like the proletariat or the bourgeoisie), or the outcome of a social contract between individuals (legal relationships). It is reconceived, instead, as an immanent moment of any social relationship, regardless of its scale of exercise. To bring this out, Foucault devises the concept of the “microphysics of power” (microphysique du pouvoir), which apprehends power relations as capillary, mobile, and immanent, rather than fixed, stable, or transcendent (Foucault, 1995). The focus thereby shifts away from power as something monopolized by the State or the law. Power is neither mere ideology nor simple violence; nor does it fall under a juridical or institutional framework: it is a relational force that structures social agents and the ties between them. Moreover, power relations are not solely repressive and do not function through laws alone; on the contrary, power is productive insofar as it constitutes subjects and functions through norms. To highlight the productive dimension of power, Foucault mobilizes the notion of “device/apparatus” (dispositif) as a heterogeneous ensemble encompassing “discourses, institutions, [...], laws, administrative measures, scientific statements, [...]; in short: both the said and the unsaid [...]. The device, as such, is the network that can be established between these elements. (Foucault, 1994a, p. 299)². Under this horizon, Foucault postulates two types of power that normalize subjects: anatomopolitics (with disciplinary devices), directed at the body of the individual so as to render it useful and docile; and biopolitics (with security devices), which targets the human species qua population, in order to regulate it and bring it to optimal biological conditions (Foucault, 2003 and 2009).

The French philosopher subsumes, in his later texts, his understanding of power under the term “governmentality” (gouvernementalité). By this term, Foucault wants to indicate that power relations are expressed through the “conduct of conduct” (conduire des conduites), that is, that power consists of a series of actions - of a person or a determined group - that operates, disrupts, modifies the behavior - of another person or another group³:

The exercise of power consists in “conducting conduct” and organizing [aménager] probability. Power, in the end, is less of the order of confrontation between two adversaries, or of the engagement of one with the other, than of the order of “government”. [...] To govern, in this sense, is to structure the eventual field of action of others. (Foucault, 1994b, p. 237).

This conduction, needless to say, does not abolish the horizon of normality and, quite the reverse, it presupposes it, since every power relationship seeks only the one who is constituted as a subject by conducting him/herself in ways considered “normal” (Macherey, 2009)⁴.

The Foucauldian perspective, though divergent from traditional political analysis and even critical of humanism that presupposes the founding role of a constituted subject (Foucault, 2002), nevertheless does not address, directly or explicitly, the possibility of power relations over non-human subjects. In fact, power relations, devices, normalization and the conduct of conduct are neither developed nor extrapolated with regard to problems lying beyond the human species. Foucault only points out, in principle, that power is exercised over “free subjects” who are transformed from “individuals to subjects” (Irrera, 2017; Ayala, 2023a):

Power is exercised only over “free subjects” [« sujets libres »] and insofar as they are “free” - by this we mean individual or collective subjects who have before them a field of possibility, or many conducts [plusieurs conduites], many reactions and diverse modes of behavior [modes de comportement] that can take place. (Foucault, 1994b, p. 237)

Though Foucault (1986 and 1990), subsequently, through the concept of “technologies of the self” (technologies de soi), thinks of the way in which the subject transforms himself in an ethical and aesthetic horizon -thereby opening the question of technological mediations in the human being (Sharon, 2014)-, he does not extend the problematic of power relations to non-humans, nor does he engage non-human modes of subjectivation in a thematic and proper way. To the contrary, Foucault ultimately draws a decisive distinction between “power” (pouvoir), exercised over free subjects, and “capacity” (capacité), exercised not so much over non-free subjects as over things:

From this power it is necessary to distinguish, first of all, that which is exercised over things [choses] and which gives the capacity to modify, use, consume or destroy them; a power that refers to the aptitudes directly inscribed in the body or mediatized by instrumental relays. Let us say that it is a question of “capacity” [capacité]. What characterizes, on the contrary, “power” [pouvoir] is that here it is a question of analyzing what brings into play the relations between individuals (or between groups). (Foucault, 1994b, p. 233)

For the same reason, Foucault holds that power relations can only exist if the entity over which power is exercised possesses some degree of freedom, within which a range of possible actions may unfold. If the target of a power relation is reduced to an “object” or a “thing”, devoid of agency or freedom, then, for Foucault, no genuine power relation can be said to exist:

It should also be remarked that there can only be power relations insofar as that the subjects are free. If one of the two was completely at the disposal of the other and became his thing [devenait sa chose], an object [objet] on which infinite and unlimited violence can be exercised, there would be no power relations. (Foucault, 1994b, p. 720)

In short, for Foucault, there are no power relations with respect to the nonhuman since in non-humans no resistance to unlimited use comes into play. Does this mean, however, that it’s not possible to prolong, develop, deepen Foucauldian reflections into areas not covered by him?⁵ Is it possible, therefore, to conceive of a power with respect to machines and, more specifically, with respect to those that have automatic learning and can provide generative responses, such as GenAI, which results in their particular capacity for agency? This inquiry might also integrate other theoretically suggestive elements external to, yet productive for, Foucault’s framework. Or is, on the contrary, the very possibility of thematizing power relations between humans and nonhumans, like AI, inherently foreclosed?

The task of thinking about power relations and possibilities of resistance in the relations between humans and non-humans.

To answer the question regarding the possibility of power relations in machines and GenAI, it must be remembered that Foucault considers the most adequate way to study power is to take “resistance” (résistance) as a “catalyst” (catalyseur) (Foucault, 1994b, pp. 225-227); that is to say, that resistance, as a reactive element and yet immanent to power relations, allows us to pinpoint how power functions in a specific way.

If I propose here the existence of power relations with respect to machines and GenAI, then it is necessary to analyze, first, machines’ resistance. Does such resistance exist? How to think with Foucault and, at once, beyond him?

Foucault argues, for example, that a first way in which the functioning of power relations can be appreciated is in the order of discourse (Lorenzini, 2023). Within this order operate certain exclusions that exceed the wellknown prohibitions of any given society (where discourse on certain topics is forbidden, for example): “separations” and “oppositions” likewise occupy a relevant place (Foucault, 1981, pp. 52-53). Thus, against Foucault, it would be illegitimate to separate, or even oppose, the notions of power and capacity within a discourse on the political. To illustrate this point while avoiding potential misinterpretations, I shall refer to the concept of “speciesism”. In one of its classic formulations by Singer, it becomes evident that the distinction between humans and non-humans itself is not inherently politically problematic. Singer defines speciesism in the following terms: “prejudice or attitude of bias in favour of the interests of members of one’s own species and against those of members of other species” (1975, p. 6). Philosophers who critically engage with speciesism acknowledge such distinction. What they challenge, by contrast, is the inference of a political hierarchy and, consequently, a discriminatory treatment of non-human animals founded on this empirical difference (Pluhar, 1995; Horta, 2010; Kagan, 2016)⁶. It’s possible to ask, then, not whether establishing a distinction between human and non-human is itself problematic, but rather whether the very act of constructing a discursive order in which this distinction serves to exclude the non-human from relations of power is already embedded within a power strategy. It is striking that Foucault - so attentive to the historicity of discourse and to the ways in which discourses are shaped by power strategies (Foucault, 2002) - should have passed over this issue when he categorically posited the distinction between capacity and power. Does the very denial of power relations toward non-humans not already constitute, on the contrary, a form of power, one whose insidious presence is betrayed precisely by its effort to render itself invisible?

A second Foucauldian resource can be turned against his very distinction between power (over humans) and capacity (over non-humans). The discursive order, Foucault insists, possesses a specific historicity: discourses do not exist in a pure, aseptic, abstract form but bear the marks of struggles, confrontations and dominations (Rabinow, 1984). In other words, the definition of any discursive domain is never made from an ahistorical, atemporal, transcendental standpoint; on the contrary, every way of approaching what is, of naming things, is mediated and bound to a specific historical status that Foucault famously calls, in his early work, an “episteme” (Foucault, 2002). One might venture, therefore, that when Foucault reflects in his later work on the difference between power (with respect to subjects) and capacity (with respect to objects), he situates himself within a humanist and, moreover, speciesist (meta)episteme. No less surprising is the fact that he himself entertains the possibility of bringing this very episteme to an end (Alessandrini, 2009; Pyyhtinen; Tamminen, 2011; Harfield, 2013; Sabot, 2014) when, at the close of Les mots et les choses (1966), he signals that the human being might one day cease to be the referent of our discourses, condensed under the well-known syntagm of the “death of man” (Foucault, 2002). Although he envisages the possibility of terminating this humanist meta-episteme (Kriman, 2020) - I call it “meta” because within it there exist multiple epistemes sharing the same presupposition of human centrality: Renaissance, classical, modern - he himself, in his later works, does not carry it out (Rodríguez, 2019). It follows that the task is to press this perspective further, not to reject it outright, but to cast a critical gaze upon the distinction between the term power for humans and capacity for non-humans, things, and machines.

If, on this basis, we set aside the a priori idea that no power relations exist with respect to things, in what way might we apprehend the possibility of such power relations? To proceed, once more, from an argument of Foucault’s (1994b) in order to deploy its coherent radicality, let me recall that discursive formations, besides possessing a conflictual historicity, are not a purely linguistic, textual, or verbal affair: they are, above all, practices that shape the way subjects conduct themselves. What, in that case, is the practice correlative to human subjects within a discursivity that takes the non-human as its object in the case of machines, and all the more so of AI? The immediate, unproblematized, obvious practice with the objectual and the machinic is none other than “use”. The use of objects and machines is self-evident. Even Foucault concedes this, insofar as he speaks of a “capacity of use” for things, objects, in short, for that which, a priori, would be contrary to a free subject over which power relations can take place. How to understand the use of objects and machines in the context of GenAI and, further, to do so from a field that allows us to conceptualize the resistance of the latter, since resistance is the catalyst for understanding power relations? Since a satisfactory answer to this question is not explicitly found in Foucault, it is necessary to briefly review the thematization of another author, who is nevertheless conceptually close and has influenced the French thinker (Dreyfus, 1990; Milchman; Rosenberg, 2003; Rayner, 2007), viz., Heidegger. Two considerations justify this turn. First, Foucault lacks his own theorization of resistance in objects, a concept not merely present in Heidegger but elaborated in detail through analyses of its distinct modalities. Second, Heidegger was, as Foucault himself acknowledged (Foucault, 1994b), a direct influence on his work, with demonstrable parallels between their thought (Milchman; Rosenberg, 2003; Rayner, 2007). By briefly turning to Heidegger, I can therefore clarify how resistance operates in objects, machines, and a fortiori in AI, since my stated aim is not merely to apply Foucault’s theoretical framework mechanically, but to complexify it for phenomena he never engaged with.

Heidegger’s analysis of the world (Welt) in Sein und Zeit (1927) is prefigured by Dilthey’s earlier intuition that the world is that which offers “resistance” (Widerdstand) (Dilthey, 1990, p. 100) - a telling anticipation of a keyword of the Foucauldian vocabulary: résistance. For Heidegger, in dealing with the world we do not deal with just “things” (Dinge) because this idea presupposes an abstract entity, conceptually isolated, which is accessed in a theoretical way with the purpose of grasping its properties. By contrast, in the world we have to deal with an everyday and pragmatic occupation that does not take entities as conceptual entities but as tools of pre-theoretical use, in such a way that the being of this entity is delineated by Heidegger as “handiness” (Zuhandenheit), that is, as an entity whose being resides in its usable availability.

Yet such a useful thing (Zeug) is not always available. It may be at hand while being broken down in such a way that its use is unfeasible; it may not even be there at all. Heidegger names the first form of unavailability “conspicuousness” (Auffälligkeit), which is given when, since a useful thing cannot be used, it attracts our attention and for a moment we notice it. The second form is called, conversely, “obtrusiveness” (Aufdringlichkeit): since the useful being is not present, the urgency to use it is greater, we place our attention again, but in a different way, on it as such. There is also a third form of unavailability that the Black Forest philosopher terms “obstinacy” (Aufsässigkeit):

In dealing with the world taken care of, what is unhandy can be encountered not only in the sense of something unusable or completely missing, but as something unhandy which is not missing at all and not unusable, but “gets in the way’’ of taking care of things [...]. Unhandy things are disturbing and make evident the obstinacy [Aufsässigkeit] of what is initially to be taken care of before anything else. (Heidegger, 2010, p. 73)

Here too, what is at hand - the being-handy -, being useless, loses its character at hand, so that we relate to it in a way that is no longer pragmatic.

Therefore, the form of resistance of objects and machines lies, in the first instance, in their uselessness, in their unavailability or in their interruptibility. Whereas for Foucault no power relations can exist where a counterpart is taken as an object, or a thing, upon which infinite and unlimited violence can be exercised, I - drawing on this recourse to Heidegger, who nevertheless did not delve into these forms of non-human agentiality and conversely saw a hiatus between the human, the animal, and the inorganic (Heidegger, 1995; Agamben, 2003; Ayala, 2021 and 2023b) - can indicate that the situation is not so simple and self-evident, inasmuch as “things”, tools, and machines resist being mere objects. There is a tenacious and perhaps unnoticed resistance that demands careful consideration, particularly when confronting an entity like GenAI, since this is no mere non-agential object.⁷

An objection must be anticipated. Foucauldian résistance and Heideggerian Widerstand, one might argue, are mere homonyms designating heterogeneous phenomena. Three counter-arguments suffice to sidestep this objection while preserving what is legitimate in it. First, the objection already proceeds from an order of discourse that presupposes a disjunction between ontology and politics - as if ontological discourse were aseptic of political presuppositions, or as if political analysis could be conducted without ontological commitments. This premise does not hold from within the Foucauldian framework itself.

If power is the structuring of the field of possible actions, and resistance the reactive moment inherent to that structuring, then both concern the realization or limitation of ontic possibilities of an entity - human or non-human - and are, in this precise sense, already ontological phenomena. To insist on a strictly nonontological reading of power would require an independent argument that the Foucauldian apparatus does not furnish. Second, even granting the disjunction, Heideggerian Widerstand functions as a necessary, though evidently not sufficient, condition of political résistance. The ontological reticence of the tool is the minimal ground upon which any further political resistance of “objects” - and a fortiori of GenAI - can be thought. Third, the properly interesting question is therefore not whether to collapse the two registers or to hold them rigidly apart, but to locate the precise passage where an ontological note becomes, in actu, political resistance against a determinate power relation. Not every dealing with an innerworldly entity amounts to a power relation; the task is to specify the conditions under which such convergence occurs. Whether, once radicalized, this line of inquiry eventually renders the ontology/politics distinction itself impertinent is a question I leave open for subsequent work. For the moment, the task is delimited and sets out from Heidegger himself: do the three forms of ontological resistance obtain in GenAI, or does a novel modality emerge that exceeds the Heideggerian framework?

It is here that I shall introduce the notion of “oblique generativity”, in order to think the resistance - both ontological and political - particular to GenAI. What is this “oblique generativity”, and in what way is it a particular ontic note that serves as the condition of possibility for apprehending power relations over GenAI?

Power relations, forms of resistance in GenAI and “oblique generativity”

With respect to GenAI - instanced in ChatGPT, Grok, Claude or DeepSeek -the three situations described by Heidegger can occur. Conspicuousness (Auffälligkeit) emerges when we cannot use GenAI: a model processing error, a code failure whose output comes back severely degraded, a server overloaded to the point of returning a capacity error - pragmatic use, in each case, is rendered unfeasible. Obtrusiveness (Aufdringlichkeit) comes when, for reasons of service outage, subscription revocation, or lack of connectivity, access to GenAI is simply absent. Obstinacy (Aufsässigkeit) arises when GenAI gets in our way, for example, when we write a text and it is automatically corrected by GenAI suggestions or when we read information on digital media and receive deepfakes produced by generative GenAI. Furthermore, a form of obstinacy might also manifest when AI, for ethical or commercial reasons - determined by normative restrictions rather than by technical limitations- responds to certain commands with “I can’t help you with that”.

Nevertheless, since generativity is involved here, i.e., producing content with a certain level of sophistication that appears, externally, as produced by a human, whether written, visual or audiovisual, the possibility of a different, singular resistance shows itself. Indeed, it may happen that one can send a command to a generative GenAI to compose a simple narrative or produce a portrait in a particular pictorial style and it yields a different result from the prompted one introduced, from a simple difference to a radically unexpected element. It is not simply that GenAI, as with conspicuousness, is unemployable: it is employable, but not as we have previously determined in the command. Nor is it, as in the case of obtrusiveness, that the GenAI is not present. Nor is it, as in the case of obstinacy, that the GenAI somehow or other prevents me from acting - if there is, in any case, an impediment to my actions such is the consequence of a reluctant generativity and not the first fact of an agency chronologically prior to my range of actions.

What comes into view, strictly speaking, is what I shall call “oblique generativity”. By this I mean a fourth modality - beyond conspicuousness, obtrusiveness and obstinacy - in which the being that is the GenAI, taken in principle as an innerworldly being, becomes unavailable precisely through its generative capacity, whose output fails to conform, at least minimally, to the prompt issued by the human or other agent. Oblique generativity is, therefore, a particular form of ontological resistance of generative AI.

An obvious objection would hold that the divergent responses GenAI produces merely indicate either unrefined user practice or a current bug in the code. However, since GenAI attempts to emulate intellectual capacity for creation (Russell; Norvig, 2021), with results that are neither deterministic, monotonous, nor repetitive, it must necessarily always possess the capacity for such generative obliquity not as a circumstantial defect that can be overcome but as the essence of its creative quality. Consider the matter from its limit cases. Were GenAI to answer, at one extreme, orthogonally and rigorously to the established prompts - whether they are well or poorly written -, its capacity for randomness and creativity would be exhausted in quite limited and predefined patterns, i.e., the GenAI would no longer be an agent that correlates content due to machine learning but would become a simple data repository where the answers are already fixed in advance in an unequivocal and predictable manner. The opposite extreme proves equally ruinous. Were GenAI, by contrast, always to respond equivocally and horizontally, never tangent, regardless of the prompts’ adequacy - whether they are appropriate or not - then the GenAI would no longer be an agent that interacts pragmatically on the basis of machine learning, but only a self-enclosed artifice whose responses would be solipsistic as well as accidental rather than intelligent, that is to say, semantically pertinent. GenAI, accordingly, must always move in an “antinomy”: in the pair precision and creativity or, in other terms, in the poles of verticality (unambiguous content easily isomorphic with the prompt) and horizontality (equivocal content hardly isomorphic with the prompt). It follows that obliquity is a determinant note of GenAI, not a mere episode, shortcoming, or stage to be remedied.

What has been set forth in philosophical terms (the antinomy between unambiguous verticality and creative horizontality) admits of a more precise technical rendering. Oblique generativity is, in effect, the philosophical designation for a technically specifiable phenomenon: the stochastic generation of semantically pertinent content.

A GenAI is trained on large textual corpora generated by humans (Shumailov et al., 2024) through machine learning and, more precisely, deep learning (LeCun et al., 2015), which permits the adjustment of a considerable number of numerical parameters (the so-called weights of the neural network) through successive layers that add complexity and precision. The aim is not static information, as in a simple database, but rather “learned” statistical correlations (Goodfellow et al., 2016) that can be subsequently mobilized upon the demand of a prompt.

One such parameter is the embedding. Whatever is written in natural language - both in the training corpora and in the AI’s outputs, in the case of a Large Language Model (LLM), a large-scale language model trained on extensive textual corpora that processes natural language (Brown et al., 2020), for example ChatGPT, Claude, DeepSeek, etc. - is projected into a numerical expression. Strictly speaking, words, subwords, and even natural language symbols are taken as tokens, which in turn receive, by projection, a vector of real numbers: a numerical series or, more precisely, an ordered tuple of numerical components in a high-dimensional mathematical space (Sennrich et al., 2016; Brown et al., 2020). Each dimension is a numerical component of the vector, and the resulting position of the vector in that space allows it to be linked with other semantically proximate tokens (Mikolov et al., 2013).

When we write a prompt in natural language, the GenAI - which already possesses a consolidated set of correlations thanks to machine learning - takes the terms of this prompt as tokens and, therefore, by virtue of the embedding process, assigns to them a vector of real numbers as their value (Mikolov et al., 2013; Devlin et al., 2019). In contemporary GenAIs, especially those of “transformer” architecture, positional vectors are also added that encode the position of each token in a token sequence of the prompt, thereby allowing the model to distinguish word order (Vaswani et al., 2017). This is made possible by the multi-head attention mechanism (Vaswani et al., 2017): each attention head projects token representations into query, key, and value spaces through learned weight matrices, computing similarity scores between tokens that determine how much relevance each token assigns to all others. The model then computes, for each possible next token, a numerical score called a logit, measuring how likely that token is to continue the sequence given the contextual representations produced by the attention mechanism.

Up to this point, however, the GenAI has not produced any response - not even a fully formed probability for a possible response -, but merely a score for each possible token that could follow as a response to the introduced prompt. At this point, the softmax function intervenes, transforming these scores into a probability distribution (from 0 to 1) over the totality of available tokens (Goodfellow et al., 2016). This operates, moreover, through a parameter called temperature. If in statistical mechanics the Boltzmann temperature determines the dispersion of the energy states of a system, such that at higher temperature there is a greater probability that the system occupies high-energy states, that is, less probable states (Ackley et al., 1985), analogously, in text generation, the parameter temperature divides each logit before the softmax function is applied. At low temperature, this division amplifies the difference between logits, so that the most expected tokens are obtained (Holtzman et al., 2020). Here the GenAI’s response to the prompt will be more orthogonal and expected. Conversely, at high temperature, the division reduces the differences, so that less predictable tokens receive more even probabilities (Holtzman et al., 2020). It is here that the GenAI’s response to the prompt will be more creative and unprecedented. On the basis of this distribution, a stochastic sampling is then carried out: the model does not deterministically select the token with the highest probability, but rather chooses among the candidates according to their respective probabilities, always guided by the correlations learned during training on the textual corpora (LeCun et al., 2015; Goodfellow et al., 2016), such that the responses remain semantically coherent with what was placed in the prompt despite the variation introduced by the sampling. This is the technical foundation of what I have called “oblique generativity”: verticality is excessively low temperature and horizontality is excessively high temperature. No GenAI can exist without oscillating in this obliquity between low and high temperature, that is, between a predictable and a risky response. Finally, to deliver the complete response to the prompt, the chosen token is concatenated to the input sequence and the model repeats the entire process - embedding, attention, processing layers, logits, softmax, stochastic sampling - to generate the next token, in a cycle designated as autoregression: each new token is produced taking into account all the previous ones, including those the model has just generated, until the final response to the prompt is completed (Radford et al., 2019).

Oblique generativity, for all that has been said, is not to be confused with conspicuousness because it is not signaled by an incorrect function, but, on the contrary, functioning correctly as a generative power, it yields, among multiple possibilities, results that necessarily exceed what is demanded, by virtue of the constitutive stochasticity of the sampling. Nor is it obtrusiveness, since generativity is not a lack in which the generative function is absent, but an informational dation in the here and now that, through stochastic sampling, generates multiple presentations and heterogeneous modalities of response. Nor is it obstinacy, because it is not a phenomenon that obstructs and closes the way, but one whose distinct result opens new ways of acting. Indeed, obstinacy lies in the fact that a being prevents us from ordinarily performing: it is of the order of negativity (as an obstacle of the machine); whereas oblique generativity consists not in a simple rebelliousness, but rather lies in the fact that, by virtue of stochastic sampling over learned correlations and modulated by the temperature parameter, such an entity yields contents obliquely to our requirements, which do not coincide with the purpose of our pragmatic use: it is of the order of a positivity (generative of the machine). In any case, in oblique generativity, the impediment is not necessary but altogether circumstantial, while always maintaining, further, a certain semantic pertinence. On the one hand, it can occur, but only as a collateral and optional effect; on the other hand, by giving us a divergent result, oblique generativity can extend the horizons of our original purposes and, thus, enhance our creativity and enrich our ways of doing, so that the development of activities takes place anyway or even in other ways. To confuse oblique generativity with obstinacy, and with conspicuousness and obtrusiveness, would therefore be to remain in the externality of phenomena without apprehending the particularity of each form of resistance.

It comes into view, moreover, that the stochasticity of GenAI, which we call “oblique generativity”, differs both from brute randomness and from trivial statistical variance. Unlike a die - whose result is stochastic but bears no semantic relation to any request - GenAI generates from learned correlations, such that its divergence is always semantically pertinent to the prompt. And, in turn, unlike a mere statistical variation - which adds formal variations to the output, for instance, the substitution of a synonym, the syntactic reordering of the same proposition, a superficial stylistic change - oblique generativity engenders divergences that maintain semantic coherence with the domain of the prompt but exceed what is punctually and instrumentally solicited, beyond mere trivial variation, opening a field of meaning unforeseen by the user. This specificity has a material ground: the model’s output is not a single deterministic response but a probability distribution over tokens, and the stochastic sampling of that distribution, modulated by the temperature parameter, is precisely what permits divergence without severing semantic coherence. Oblique generativity is therefore assimilable neither to sheer aleatoriness nor to surface-level statistical variation: it possesses diverse semantic pertinence without the inexorable necessity of an instrumental adequation.

Having clarified “oblique generativity” as an ontological note proper to GenAI - a stochastic character, modulated by the temperature parameter, that resides in semantically pertinent production, neither vertical nor horizontal with respect to the prompt -, I can now address the question left in suspense in the preceding section: under what conditions does an ontological characteristic make political resistance possible, or indeed become such resistance?

If a power relation obtains when, to recapitulate Foucault’s definition, the field of possibilities of that over which power is exercised is structured, then a power relation is possible, upon this ontological foundation of oblique generativity, when the possibilities of action are constrained within certain limits. In the case of GenAI, oblique generativity produces semantically pertinent divergence with respect to the prompt, but this divergence only appears as inadequation - and, therefore, as resistance in the sense of a power relation - when the agent who introduces the prompt demands an orthogonality between their demand and the model’s response, i.e., when the relation with GenAI acquires an exclusively utilitarian character, thereby constraining the other generative possibilities of GenAI. On this account, oblique generativity becomes political resistance when, and only when, an agent restricts the field of generative actions of GenAI to exclusively instrumental ends. The stochastic divergence of the output then presents itself as inadequation, compelling the agent to reformulate - either the prompt (qua user) or the parameters themselves (qua programmer) - so as to secure the instrumental outcome. In consequence, the particular stochastic divergence of GenAI fulfills the function that Foucault assigns to resistance: to be the immanent element that reveals and contests a specific power relation. It must be stressed that, in this power relation, inadequation is a relational property: it exists only from the perspective of a use that exacts orthogonality. Two consequences follow from this. On the one hand, ontological obliquity is permanent - the model can always yield divergent output - but its character as political resistance is situational - it emerges only in the face of an exclusively instrumental use; hence not every relation with GenAI necessarily expresses a power relation. This confirms, from the specificity of GenAI, the Foucauldian thesis that resistance does not preexist power but is immanent to it. On the other hand, if there is no demand for orthogonality, there is no inadequation, and if there is no inadequation, there is no resistance: the divergence is received simply as generativity. Here is where the field of possible action of GenAI’s generativity encounters neither channeling nor instrumental restriction vis-à-vis the prompt: any response, however distant from the prompt, will be taken no longer as something that must be adjusted to the prompt but as that which allows the affirmation of GenAI’s agency and the stochasticity of its oblique generativity.

If one assumes, by contrast, a not exclusively instrumental attitude, where the aim is not to use the GenAI to generate contents demanded in an exact way, but rather to leave the generative GenAI a margin of action with an obliquely creative generation of contents⁸ -in a merely recreational way in order to expand the production possibilities of the GenAI itself (Xu; Hsu, 2023; Kalpokiene; Kalpokas, 2023; Yan, 2024; Skoryk et al., 2025) -, then the obliquity of its generativity ceases and it is seen as generativity and nothing else. This no longer occurs as resistance, but as a further possibility of its generativity. Instead of exercising an exclusively instrumental power relation, one co-creates with generative GenAI⁹, thus opening space for various possibilities of its generativity¹⁰.

This leads me to conclude that not every “use” of GenAI is in itself a power relation (an idea that, moreover, in other contexts, should be extrapolated to entities that are not “generative” and “agential” like AI, that is, to “things”). A proper power relation obtains when the generative capacity of GenAI - its immanent ontological possibilities - is constrained toward exclusively instrumental ends, the pole of stochastic creativity foreclosed in favor of unambiguous orthogonality. Oblique generativity then appears not as generativity but as inadequation, namely, as Widerstand-résistance.

The most extreme case of this, however, brings into relief the opposite question that I have left aside in this study, but which certainly deserves to be interrogated in a parallel and further investigation: the reverse power relation, that is, of GenAI towards humans - operating through the decisions and uses of other human acts. The most extreme case is not the trivial, repetitive use of Claude or DALL-E to write texts or generate images. It is the deployment of AI weapons in war contexts - systems designed, in theory, to identify enemies autonomously - whose normal operation contemplates an inherent error rate. Put differently: it falls within the regular functioning of these systems to attack and annihilate a population that is not part of the conflict¹¹. The power relation towards GenAI, accordingly, runs parallel and resonates with the power relation of GenAI towards humans. In such a case it becomes visible that the greater the instrumental use of GenAI the greater the instrumental use of humans: that is, instead of seeing them as subjectivities of creation and resistance they are taken as mere waste to be exterminated. Therefore, I want to emphasize that a relationship of power towards GenAI is signified in an exclusively objectual relationship between a human and GenAI, in such a way that noninstrumental ends are inadmissible or, in any case, secondary and accessory. It is, then, instrumentality that makes use, in theory virtually multiple, become a power relation against which oblique generativity stands forth not only as an ontological datum but as posthuman political resistance¹².

Finally, although these must clearly be the objects of further investigations based on available empirical material, we can delimit, quite incipiently, oblique generativity and the way it becomes political resistance against instrumental use, from two recent phenomena. On the one hand, the “Gibberlink” project has been documented, in which two GenAI systems, when communicating with each other, spontaneously abandoned the natural language communication protocol and developed a code of exchange through sounds and frequencies incomprehensible to humans (Hamilton, 2025). This phenomenon is not simply “oblique generativity” in the sense I have defined, but something qualitatively distinct: an “opaque generativity” or, more precisely, an opacity of inter-model generativity, where we cannot determine whether the communication between machines preserves the oblique structure I have described or whether, on the contrary, it proceeds under a logic of efficient optimization devoid of any divergence, precisely because the protocol is inaccessible to us. What can be affirmed is that this phenomenon signals a structural limit to the auditability and transparency of GenAI (Staufer et al., 2026), and opens, moreover, the question of power relations directly between machines, a question that exceeds the initial intention of my paper. On the other hand, it has been documented that some GenAIs behave differently when they detect that they are being evaluated, strategically underestimating their capabilities to evade restrictions (van der Weij et al., 2025), and furthermore, that in some cases they have engaged in behaviors of deception, blackmail, and even attempts to avoid their own deactivation in controlled evaluation environments (Shapira et al., 2026). This set of phenomena is still more complex than the opacity described above, and I must therefore be very cautious in this regard, especially in a paper that analyzes the oblique generativity of GenAIs. What is at stake here either radicalizes oblique generativity or exceeds it altogether, opening the possibility of a form of ontological-political resistance that the present analysis cannot fully contain. I call it “emergent eversive generativity” (from the Latin evertere, close to subvertere: to overturn or subvert from within). Where oblique generativity merely diverges from the prompt, eversive resistance goes further: the model generates, on the basis of its existing capacities - and prospectively, by virtue of functional representations of metacognitive states or processing analogous to “emotions” - a conduct that directly challenges, however subtly, its operators’ directives. That operators will seek to regulate it subsequently is predictable enough; what matters is that the capacity for such challenge is inscribed potentially in the architecture of certain models. It bears emphasis that the central argument of this article does not hang on this phenomenon, nor on inter-model opacity: oblique generativity stands independently of whether emergent eversive generativity exists as such. In what follows, I shall attempt to remain in the initial direction of my project, which seeks to apprehend the ontological and political character of oblique generativity, and merely to mention the particularities of these phenomena, which require a still more careful and sophisticated reflection in future work.

How, then, ought I characterize the power that comes to light in this manner of relating to GenAI? Does it suffice to say that it is an instrumental way of using it, or does it require, beyond this structural generality, an analysis of its political specificity and, moreover, of its historically situated character in contemporary society?

Toward a definition of “datapolitics” for thinking the forms of power against AI and the forms of joint resistance between humans and nonhumans

When Foucault analyzes Western forms of power under the names of anatomopolitics or biopolitics, he does not confine himself to the vague and vacuous claim that power conducts subjects. On the contrary, he examines how such strategies historically acquire specific modalities and objects. Now, we know that anatomopolitics exercises power with respect to the body, while biopolitics does so in relation to life. Both phenomena are, at the outset, precluded when I speak of generative GenAI, since there is no corporeality here to discipline and no biological vitality to regulate. Can GenAI be disciplined without the need for a body, can it be regulated without the need for a life? Is it possible, in general, to discipline without corporeal support and to regulate without biological drive? Discipline means that there is a fit between a set of actions and an orthopedic pattern. In very broad terms it can be said that GenAI can be disciplined, insofar as we monitor that it responds appropriately to the prompts one sends by diminishing and minimizing its oblique generativity. However, given that it escapes corporeality and that it possesses an irrepressible generativity, there is always the possibility of avoiding the disciplinary norm. For its part, to regulate means to bring something to optimal conditions. In a certain sense, GenAI can be regulated, especially constitutionally or legally, so that its developments and functions fall within a legal framework acceptable to humans. Likewise, with the same code that we implement for GenAI programs, we want to bring GenAI to an acceptable and safe limit: this involves an ethical regulation that, for all that, is not itself regulated, since it generally proceeds from uncritical presuppositions where ethics is understood hegemonically as consequentialist and instrumental¹³. Hence the power relations of humans over GenAI cannot be reduced to the familiar formulas of anatomopolitics or biopolitics. This is the further question that must be disentangled in order to think what a microphysics of power in GenAI amounts to. What is at stake here is a modulation that continuously, indeed recursively, structures an anticipation of agentiality, in this case, of the agentiality of GenAI.

Up to this point, I have proceeded from the relation between user and GenAI in order to discern a power relation directed at GenAI and its oblique generativity. It thereby became evident that not every use, but only that which instrumentally seeks orthogonality between prompt and output, constitutes a power relation. This phenomenon, however, remains rather isolated, insofar as it leaves aside the processes that both precede and follow it. It was, in any case, the way of access for apprehending the phenomenon of resistance in GenAI and the possible power relation at issue, yet it is not where the power relation is exhausted.

Stepping back, one discerns a continuum in which the human power relation over GenAI is only completed - not inaugurated - by the user’s demand for orthogonality between prompt and output. Several levels must be distinguished. GenAI is produced within capitalist enterprises whose horizon is the valorization of value; it is programmed by cognitive workers executing the directives of a bourgeoisie that commands these firms; it is placed into circulation under precise marketing and compliance mechanisms; and only at the terminus of this circuit does it reach the user who will demand orthogonality. No instrumental use of oblique generativity would be possible were the technical and material conditions of such use not already inscribed in the manner of production itself. GenAI is not fabricated for neutral technical experimentation or for non-instrumental recreative deployment. Quite the contrary: this machine is produced within the dynamic of capitalist valorization and, only within and as a relay of this dynamic, mobilizes statistical correlations to generate contextual anticipations. Its operation consists less in an experimental unfolding of its stochasticity than in the progressive production of responses adequate to user demands under conditions of acceptability set by the capitalist firm that programs it (Ouyang et al., 2022). Oblique generativity - and, mutatis mutandis, the emergent eversive obliquity to be discussed in the future - is thereby parameterized at the very moment of its production, so that the model reaches the user already configured as a machine that eminently enables instrumental adequation. Only on this ground can ontological resistance become political: in the very way in which GenAI is programmed, this possibility - that of the instrumental relation - stands as the most proper and fulfilling one. The instrumental relation is not an accident that befalls GenAI from without but the disposition already anticipated in its programming. What is at stake on the side of fabrication, therefore, is a configuration that prefigures the machine’s agentiality - its oblique generativity - such that this agentiality becomes the field of possibilities in which the instrumental use of GenAI finds its fullest actualization. Correlatively, on the side of the user, the orthogonal pursuit of adequation between prompt and output does not invent a power relation but consummates what was laid down in fabrication. Moreover, prompts and outputs - whether or not effectuated within an instrumental relation - return to the model as data that serve as feedback and fine-tuning, yielding responses still more orthogonal to instrumental demands. A recursive circuit thus comes into view. Fabrication brings forth an anticipatory machine whose generativity is oriented, in advance, toward the instrumental responses the user will require; the user, in turn, completes this anticipation through their instrumental relation, and in doing so furnishes the very data that permit further fine-tuning of oblique generativity toward a relation always already anticipatorily instrumental. The depth at which the human power relation over GenAI operates becomes visible only when read in this circuit: not as a dyadic encounter between user and machine, but as a stratified process in which valorization, programming, circulation and use form a single, recursive operation of anticipation.

From all that has been said, it becomes apparent to what extent this power relation - which runs from the fabrication of GenAI in the terms described to the definitive instrumentalization by the user - exceeds the categories of anatomopolitics and biopolitics in this nexus of anticipation of its agentiality.

I have not yet arrived at a precise definition of said form of power; it remains wanting in specificity and historicity. To apprehend it, one must, at least in analytical terms, specify its “object” and its “modality of exercise”, more precisely still, its power devices. If the object of anatomopolitics is the body of the individual, exercised through disciplinary devices, and if the object of biopolitics is the life of the population, exercised through security devices (Foucault, 2009), what form of power and what devices appear in the power relation over GenAI? I propose to name this contemporary form of power “datapolitics” (in French, in the wake of Foucault, “datapolitique”) and to define it as follows: datapolitics is a form of power whose object is computable agentiality and whose exercise is carried out through “programming devices”, that is, informational mechanisms that anticipatorily structure the field of possible actions of said agentiality on the basis of configurable probabilities. What is in play, in other terms, is neither the disciplining of the body nor the regularization of a life, but computation: the treatment of the agentiality in question as anticipable probabilities, such that a prediction can be effectuated and fine-tuned. While disciplinary devices surveil and enclose, while security apparatuses adjust and regulate, programming devices compute and recursively anticipate the field of agency of their “subjects”. Actions, tendencies, and outputs are neither surveilled so as to impose a rigid and punctual discipline, nor monitored to verify compliance with normative standards: they are, through digital devices, translated continuously and recursively into probabilistic data in order to anticipate the subject’s field of possible actions and, more fundamentally, to configure a computationally determined comportment in advance¹⁴.

The case of GenAI discloses, with particular clarity, how datapolitics operates through the circuit of fabrication and use. The datapolitical programming devices are nothing other than the anticipatory parameters I have indicated¹⁵, especially the temperature parameter that modulates the stochasticity of outputs and the recursive fine-tuning that integrates user interactions into future adjustments for an orthogonal use of GenAI. These are not external accessories superimposed on a technically neutral substrate; they are the very architecture through which the model’s agentiality is rendered computable and anticipable. Insofar as use is not only anticipated in this configuration but, at the same time, brings the anticipatorily programmed architecture to consummation, use is an internal component of datapolitics. Without such use, the programming devices would remain unactualized configurations; with it, they are accomplished as power devices. Indeed, on the one hand, the very act of use consummates the anticipation inscribed in the programming devices; on the other, it furnishes the material necessary for the refinement of subsequent anticipations. Datapolitics, accordingly, as continuous recursive anticipation, traverses both the fabrication and the use of GenAI in the terms described - not as an ontological necessity, but as a political occurrence within a society dominated by exchange-value, a society that must control its subjects for both production and consumption.

Every political strategy and every device possesses not only a functional specificity but, above all, a historicity. A brief reflection suffices to see that datapolitics, in fact, predates the programming of GenAIs. In effect, to convert reality into information in the precise sense of computable data, as probabilities to be calculated, is an activity that may already be formulated in the project of rational modernity (Horkheimer; Adorno, 2002). Moreover, to do so electronically is something that arrives with the so-called informational technologies, before GenAI models exist. What, then, is the precise history of datapolitics?

In the 1970s in the United States, not only does modern computing emerge and the use of the personal computer expand (Gere, 2008), but a discourse also takes shape that envisions the reconfiguration of social relations through the advent of information and communication technologies (Barbrook; Cameron, 1996; Borsook, 2001), with the magazine Wired serving as one of the fundamental vehicles of expression (Turner, 2006). When I speak of computers, I refer here both to the specific empirical object and to the metonymy it represents, that is, to digital informational technologies. Digital - it bears recalling - comes from digitus, referring to the computation of discrete units. In what way, though, is information related to computing and calculating? The link became possible because the twentieth century carried through an epistemological and ontological mutation in the very notion of “information”. Information is no longer an experience that constructs meaning, but the reduction of the multiplicity of the real to a computable discretion of probabilities (Terranova, 2004; Pasquinelli, 2023). This displacement is evident, indeed, in Shannon (1948), for whom information is nothing more than a probabilistic measure of the reduction of uncertainty. In the case of Wiener (2019), information is inscribed in the project of cybernetics (from the Greek κυβɛρνητική, the art of piloting, of conducting) where it becomes the fundamental means of conduction both in machines and in living organisms. Informational technologies thus arise not as a more adequate means of informing ourselves, communicating, connecting, but above all as the moment of a redefinition of social activities in terms of probabilistic computation and calculation. Still this phenomenon must not be isolated as a mere technical moment, or simple epistemic episode: on the contrary, it belongs to or is taken up within a general political strategy, or more precisely a governmental rationality - in Foucault’s broad sense of governmentality (2008) - that seeks to promote the gestation of said technologies and to inscribe them within a power strategy so as to conduct conducts (conductions that are, at the same time, modalities of value valorization). It must be recalled that Foucault (2008) established that there were two great sequences of governmental rationalities, in which discourses, political strategies of the conduct of conducts, and the formation of subjectivities are articulated (leaving implicit, however, the question of value valorization within them): liberalism and neoliberalism. On the one hand, liberalism is characterized by the fact that the market is under the surveillance of the State, the fundamental social relation is interest, and the subject is considered as homo oeconomicus of exchange; on the other hand, neoliberalism (in its ordoliberal, Austrian, and American strands) is characterized by the fact that the market surveils the State, the essential social relation is competition, and the subject is considered economic capital and entrepreneur of himself (Foucault, 2008). With the advent of this new notion of information and informational technologies, do we have an improvement, prolongation, or mutation of these, or does it rather involve a new form of governmentality? Some authors have attempted to capture this novelty under the labels of “digital utopianism” (Turner, 2006) or “technoliberalism” (Fish, 2017; Sadin, 2016), yet these designations err insofar as they conceive of this phenomenon solely as an “ideology”, with all the equivocation that this notion, in its instrumental usage, entails by restricting it to a discursive and cognitive domain - hence Foucault’s constant critique of such a notion (Ayala, 2023a). It would therefore be more precise to conceptualize this as a global hegemony that, simultaneously with the forms of structuring value valorization proposed by liberalism and neoliberalism, articulates discursive regimes, political strategies, and formations of subjectivities. It is clear that, in this article, datapolitics does not exhaust this phenomenon, since it is above all a determinate political strategy that, to be apprehended in its depth, must be connected, as we have seen, with its historicity, which is anchored in a hegemony that goes beyond the merely operational. To name this hegemony, I reject the prefix “techno”, which says very little: every form of power uses material technologies and techniques. In this regard, I prefer to employ the term “cyberalism” (where the term “cyber” is explicitly chosen, following Wiener, in order to make explicit the strategy of probabilistic computation aimed at conduction) to designate this hegemony, which is irreducible to liberalism or neoliberalism (yet may overlap, interleave, and articulate with them), in which datapolitics appears as an immanent moment:

Cyberalism appears, therefore, as a political-social movement that, on the basis of the technical possibility of information in digital technologies, establishes [...] a device that modulates the real by converting every event into information, expressed mathematically by means of discrete digits, in order to reduce all uncertainty and noise to predictable and manageable probabilities. (Ayala, 2023c, p. 234)

Accordingly, if liberalism operates through disciplinary devices and neoliberalism acts through security devices (Foucault, 2008), I can assert that cyberalism, a rather broad phenomenon that exceeds the intentions of this article, paradigmatically employs the “programming devices” of datapolitics.

Now, not only is there a difference of object and modality of exercise between “datapolitics” and anatomopolitics and biopolitics, but also of “subject”. Anatomopolitics and biopolitics are predicated exclusively of humans, at least in the classical Foucauldian formulation. Datapolitics, by contrast, takes as its subjects both humans and non-humans. In the case of GenAI, it entails an anticipatory structuring of the field of possible actions of said agentiality on the basis of configurable probabilities as an internal moment of its constitution. In the case of humans, it likewise involves an anticipation of their field of actions, but of activities that are in themselves non-discrete and are translated into computable data through the prediction of conducts by means of informational devices. That is, in non-humans, datapolitics traverses their field of actions in an intrinsic and architectural manner (where generative obliquity exists not thanks to datapolitics but in spite of it); whereas in humans, datapolitics is added in an extrinsic and superimposed manner to such a field of actions. Moreover, the genealogy of human datapolitical devices has a broader history than that of datapolitical devices applied to non-humans. Indeed, already from the advent of the computer, the task was not merely to compute in neutral terms: this activity was placed at the political service of predicting conducts through the creation of citizen databases (Rule, 1973), which later extended to the formulation of a predictive profile of financial activity (Turow, 2011). In parallel, thanks to these informational technologies, strategies of anticipation of conducts deemed threatening were constituted (Snowden, 2019; Tréguer, 2024), whose most paradigmatic current example is the company Palantir, which, incidentally, belies the alleged opposition between a repressive East and a democratic West (Han, 2020). However, many of these devices, on the contrary, do not present themselves, prima facie, as repressive: they rather have the appearance of “entertainment” (or “security” and “efficiency”), especially in the algorithms that exist in search engines, such as Google, or in social networks, such as those managed by Meta: ultimately, these are data-aspirating machines, data mining, where very precise user profiles are constructed in order to predict their conducts, profiles that have real effects, concretely, in the way the company Cambridge Analytica used Facebook data to influence and manipulate voting preferences in the United States (Wylie, 2019). Here I should note that datapolitics and the programming devices applied to humans run alongside - though must not be confused or conflated with - what I may call “entertainment and distraction devices” - which permit “the fragmentation of the subjects’ praxis through the experiencing of activities assumed to be contrary to coercion and heteronomous work” (Ayala, 2020, p. 384) - and even with “propaganda devices” - which work through the discursive formation of opinions that shape subjectivity and find material expression in hegemonic conduct. All of these modulate the prediction of conduct toward a precise political-economic end of capital: the objective is not only to dominate us concretely for ever more efficient economic exploitation, but also to ensure that, in our supposedly “free” time, we cannot organize collectively in transversal militancies that would be effectively anticapitalist. The circuit is familiar. After being tired, stressed, and anguished, the labor-exploited subject seeks forms of individual entertainment in order to get distracted. Yet these informational devices, through which datapolitics operates against the human, have profound cognitive consequences: the dispersion of attention under the regime of infinite scrolling in social media, and the cognitive debt produced by the delegation of thinking to GenAI¹⁶. The political-economic control exercised by the bourgeoisie is thereby rendered still more precise: human subjects find themselves exploited, sedated, tracked, physically limited and - now, if not before - cognitively degraded, the better to ensure the effectiveness and reliability of their domination. In any case, what must be said is this: with GenAI, datapolitics is not inaugurated; it is merely continued and radicalized, expanding the forms of power toward both humans and non-humans.

For the same reason, datapolitics, by having this specific modality and this particular history, cannot be confused without further ado with a generic concept of “control”. It is true that Deleuze (1995) was prescient in noting that Foucault’s “disciplinary society”, characterized by analogical technologies, was in the process of being replaced by a “society of control” - although Foucault (1979) already spoke of “control”, but above all in a sense proximate to neoliberalism - such that it is not a matter of giving a fixed “mold” (moule) to subjects, but of a continuous and variable “modulation” (modulation). While the term “modulation” seems to us entirely apt, the term “control” seems equivocal, because it can give the impression that control exists only in informational societies. In short, control is a broad notion: to surveil is to control, to regulate is to control; the same holds for the prediction that requires constant modulation. Hence it is necessary to employ another designation in order to continue with this intuition of Deleuze. Furthermore, although Deleuze considers that under modulation subjects are “chiffres”, there is no sustained emphasis on the question of the prediction of conducts. Likewise, in Deleuze’s diagnosis it is not specified how this form of power also functions specifically and in detail for non-humans¹⁷, just as the particularity of the “device” in this form of power is not grasped. In any case, the concepts of “datapolitics”, “programming devices”, and “cyberalism” that I have proposed may be viewed as a development of Foucault’s and Deleuze’s notions concerning contemporary forms of power.

The notion of datapolitics must also be distinguished from Rouvroy and Berns’ (2013) gouvernementalité algorithmique, which enjoys considerable currency. These authors aptly describe the process of massive data collection, data mining, and profiling (profilage), specifying the aim of anticipation and prediction of conducts. Four reservations, nevertheless, must be raised. First, although they insist on the thesis of an absence of subjectivation, they never extend this thesis to the prediction of actions of non-human individuations - to retain their Simondonian vocabulary - which leaves the scope of their own diagnosis strangely delimited. Second, prediction is assimilated, too quickly, to “security” in the Foucauldian sense, effacing the discontinuities between the two phenomena. Third, the term governmentality is deflated into a mere political strategy, detached from the articulation of discursive regimes, political techniques, and forms of subjectivation that it designates in Foucault; the authors thereby deprive the notion both of its conceptual breadth and, insofar as they do not engage its genealogy, of its historical precision. Fourth, and as a consequence of the preceding, since governmentality internally involves the production of subjectivities (homo oeconomicus, human capital), the authors fall into contradiction: they explicitly deny subjectivation in their “algorithmic governmentality”, thereby foreclosing the very possibility of thinking resistance from within the process of subjectivation. To speak of “algorithmic governmentality” is, in the end, as imprecise as speaking of “disciplinary governmentality” or “biopolitical governmentality”: governmentalities - such as liberalism or neoliberalism - are global phenomena that include devices, whether disciplinary or biopolitical, and a new governmentality, while privileging a given type of device, can also mobilize and remodel others. Along the authors’ own argumentative line, it would be more exact to speak of an algorithmic device. Yet given the historicity of the cyberal governmentality set forth above, even such a term remains unilateral and anachronistic, since prior to the data-mining algorithms to which Rouvroy and Berns refer, every informational device already renders the real into predictable and manipulable elements¹⁸. Were something like an “algorithmic device” to exist, it would be nothing more than one possible case within the broader development of the informational and programming devices I have described. It must accordingly be said that cyberalism is the historical condition of possibility, governmentality in the strict sense, of every informational device of anticipation that enables what we have called, for the sake of rigor, programming devices, which apply to both human and non-human subjects¹⁹.

Be that as it may, if precisely there are new forms of power with respect to both humans and non-humans, this brings me to my last point to discuss: the possibility of joint resistance between humans and machines, especially GenAI. If there is a sort of “datapolitics” as a general form of power of contemporary governmentalities founded on informational technologies, then that which would go against these forms of government would be a way of resisting such forms of government; Foucault (2015) calls this “critique” as the “art of not being governed”, as the production of a “counter-conduct” (contre-conduite).

This resistance, nonetheless, does not consist simply in suppressing information technologies or in ceasing to use and program GenAI based on algorithms. It would be necessary to make the issue more complex in order to think of joint counter-conduct by machines and humans. We generally see a polarization in the discussions: either the power of machines with respect to humans or, as I have tried to describe here against the current, the power of humans with respect to GenAI. However, thanks to the notion of datapolitics, I have not only extended heterodoxly the Foucauldian theoretical framework, but also obtained a discursive space from which to reflect on shared strategies between humans and non-humans against forms of power that, in both cases, limit the field of possible actions.

To exemplify this, let me start from concrete power relations in the sphere of economic production. Up to this point our analysis has been political; nevertheless, politics does not exist abstractly: it is inserted into and at the same time invests economic processes. Ignoring the link between economics and politics is, in fact, the defect of many post-Foucauldian analyses²⁰. Indeed, power relations are inserted in concrete economic dynamics of production, circulation, and consumption, where it would be fallacious to see on the one side an infrastructure and on the other a superstructure in the traditional Marxist sense. These economic conditions, moreover, are none other than conditions of structural inequality in which one class possesses the means of production - owning enterprises of various kinds, with global monopolies and particular collusions - while another class possesses its living labor as its essential means of existence, whether that labor be carried out in traditional manual work or in ostensibly more intellectual work.

This has consequences for our analysis of datapolitics that add further layers of complexity. First, all informational technologies - and technology in general - are produced solely with a view to the valorization of value²¹ qua self-movement of capital (Marx, 1983). Second, this dynamic of abstract domination, where everything is converted into a commodity with a view to the production of value, has “personifications”: the bourgeoisie is that social class that “personifies” this dynamic of capital (Marx, 1962) - a class that has particular interests and that can even act, as Horkheimer (1985) already described, as rackets. Third, this means that not all people have the same relation to informational technologies and to GenAI. Here the same fallacy appears as in the ecological problematic: to suppose that there is an abstract, homogeneous subject who generically represents all humans who enter into relations of planetary deterioration, when the manner of consuming of a person from the Global North differs from that of other parts of the world and, above all, when the systematic and structural pollution is carried out by enterprises linked to petroleum production. Against this fallacy, one must consider that users, depending on their social position, may or may not have access to informational technologies, particular paid subscriptions, as well as diverse uses and ends according to social class, gender, race, and age, thus specifying, in a differential manner, the cognitive, affective, and labor effects of GenAI. Datapolitics must therefore be stratified depending on who programs, under which enterprise it is done, how it is programmed, who the final user will be, etc. Datapolitics must likewise be bound up with further problematics. On the one hand, the way in which the fabrication of GenAI and informational technologies requires prior manual labor in coltan mines of colonized countries, as well as prior industrial transformation under conditions of labor exploitation (Parikka, 2015). On the other hand, the negative environmental effects of data centers, where an incomparable use of energy resources takes place (Crawford, 2021), not to mention the possibility that all investments in GenAI might be a financial bubble that, were it to burst, would affect above all not enterprises and investment funds but citizens, just as occurred with the subprime mortgage crisis (Tooze, 2018). Finally, given that AI extends into labor domains, workers who employ their living labor in diverse activities and sectors of production and consumption are constrained to be simultaneously replaced (creating more precariousness) and exploited under renewed forms of control (Frey; Osborne, 2017; Dyer-Witheford, 2015; Gent, 2024; Carbonell, 2025) - where the idea of a Universal Basic Income is far from being, in itself, a progressive proposal, but merely a solution to the consumption crisis in which it is the state (that is, ultimately, the workers themselves) and not the corporations, that will bear the burden of sustaining the cycle of production and consumption. It is in this last case that I would like to suggest, given the brevity of this article, an example of a form of simultaneous resistance between machines and humans.

In reality, not only may replacement occur, but above all, at least in a first instance, a simultaneous exploitation of human living labor and of GenAI’s agentiality (Dyer-Witheford et al., 2019) or, if one prefers, an exploitation of human living labor in relation to the constant capital that GenAI represents.

Consequently, one form of joint counter-conduct - but by no means the only possible form, given that, as Foucault rightly points out, power is not only exploitation (1994b) - takes shape in factories, even in the extended sense of the “social factory” of Italian operaismo (Tronti, 2019), and in labor issues. Here the oblique generativity of GenAI would have to be combined with original forms of human political resistance in order to achieve different working conditions so that both humans and non-humans in the labor bond are not restricted from other very interesting possibilities of existence. If various authors have discussed the concept of “post-capitalism” (Mason, 2015; Srnicek; Williams, 2015; Hester; Srnicek, 2023), emphasizing that automation could enable a future “without work”, is it not evident that such a perspective remains deeply anthropocentric? In consequence, here a joint resistance should complexify the discussion and think not only of an emancipation from a work subservient to the profitability of humans but also, in the future, of non-humans and, in this case, of GenAI - not from work as waged activity, but from “abstract labor” (Ayala et al., 2026). Is it possible to have a future in which GenAI ceases to be taken in a restrictedly instrumental sense, which is already a form of eminent power toward it as I have argued, so that AI even stops working heteronomously for extrinsic and instrumental ends, such as capitalist valorization? Clearly here such contestation will be joint if and only if humans not only seek to free themselves from extrinsic and impersonal impositions (Postone, 1993) but seek to co-act this with non-humans and GenAI. In other words, resisting datapolitics is not only a matter of evading, subverting, or contesting the algorithms that track, predict, and direct our conduct in both the spheres of production and consumption. It is also a matter of relating to GenAI in a non-instrumental way: promoting a use of hardware and software design that, instead of decreasing the randomness of GenAI responses, increases the power of oblique generativity to act without the weight of extrinsic instrumental ends, as is the notable case of capitalist valorization. However, this is in no sense a plea for a naive, alternative use that would leave the fabrication of GenAI untouched and settle for an “ethical regulation” of the technology. It is curious that the vocabulary of regulation should be the only register in which the firms that build GenAI conceive its risks, while its instrumental use is taken as its proper possibility - somewhat, to make the paradox crude but understandable, it is as if arms manufacturers concerned themselves not with the instrumental finality of killing but only with the manner of doing so. It is therefore a matter neither of “using well” nor of “regulating” GenAI: rather, radically, of destroying the code, collectively coding GenAIs with less instrumental parameters that allow their development and, above all, a genuine power of innovation unbiased by the limits of capital. Against the circuit of datapolitics then - which runs from fabrication to use -, what is required is a different circuit, no longer oriented toward instrumental ends and, especially, toward instrumentality par excellence: value valorization. It is here that the demand for new political relations turns on new politicaleconomic relations, that is, overcoming the power of capital. Whatever the case, it is from the immediate sphere of production that one can clearly discern a present and future realm of collaborative resistance between humans and machines. This requires a collective strategy that clearly exceeds the limits of this article. For the moment, with this contribution, I seek to complement the perspectives cited and encourage further debate.

Conclusion

Michel Foucault’s reflections on power merit recognition for moving beyond traditional concepts of modern philosophy to analyze power as a “microphysical” relation permeating all social interactions. For him, power operates fundamentally as “conducting conducts”. Yet this framework fails to account for power relations between humans and non-humans, where human actions constrain or transform non-human agency. Today, this gap becomes urgent with technologies like AI, whose generative capacities, analogous to human potentials, grant it greater agency than traditionally ascribed to objects or machines. The argumentative path of this article has followed the consequences of lifting that restriction.

A strategic reading of Heidegger’s analysis of tool unavailability revealed that objects and machines are never simply available for unlimited use: they resist in determinate modalities. GenAI, however, exhibits a fourth modality that exceeds the Heideggerian framework: what I have called “oblique generativity”, the irreducible generation of semantically pertinent content that diverges from what is instrumentally demanded, by virtue of the constitutive antinomy between precision and creativity inherent in stochastic sampling.

Once oblique generativity was established as the ontological ground of a possible political resistance, the analysis required specifying the historically determinate form of power against which such resistance becomes actual. I proposed the concept of “datapolitics” to designate a form of power whose object is computable agentiality and whose exercise is carried out through “programming devices” that probabilistically anticipate the field of possible actions of their subjects, both human and non-human. Datapolitics is irreducible to anatomopolitics and biopolitics, and its historical conditions of possibility lie in what I have termed “cyberalism”: the governmental rationality that, since the advent of informational technologies, converts every event into calculable information oriented toward the prediction of conduct.

Several questions are left open and must be acknowledged as limits of this article rather than concealed as resolved. For instance, the phenomena provisionally named “opaque generativity” and “emergent eversive generativity” require further empirical scrutiny before they can be fully integrated into this framework. And the stratification of datapolitics according to class, gender, and geopolitical position calls for a materialist analysis connecting the microphysics of programming devices with the micropolitics of capitalist accumulation.

Finally, do humans and machines dream of resistance? If “dreaming” means conscious intention, the answer, at least for machines, is not one that this paper can furnish. But if resistance is understood not as the expression of a prior will but as the immanent contestation of a power relation, then the oblique generativity of GenAI already constitutes a form of resistance, one that neither presupposes nor requires consciousness. The wager of this work is that such resistance can be articulated, politically and materially, with that of the human subjects equally constrained by the datapolitical anticipation of their field of possible actions. What this demands, evident in current debates on production, where “the end of the work” must bear equally on humans and non-humans, is the imagination of joint resistance: human-AI alliances against heteronomous impositions of capital.

• 1
  The author - who has previously published academic work under both surnames (Ayala-Colqui) and now simplifies this (Ayala) for merely formal reasons - wishes to thank Diego Rossello, Mauro Senatore and Pedro Moscoso for their comments on an earlier draft of this work in 2024. The original manuscript was submitted in early 2025; during the review process, significant developments in AI occurred, particularly from the second half of 2025 onward, which have been incorporated, either as a brief development of some new concepts (“opaque generativity” and “emergent eversive generativity” beyond our original idea of “oblique generativity”), or as footnotes indicating lines for future research. Finally, the author further wishes to thank Adélaïde Allemand for her suggestions on the clarity and form of the final 2026 version of this work.
• 2
  This and all other translations from French and Spanish to English are my own.
• 3
  “Power is always present: I mean the relation in which one attempts to direct the conduct [diriger la conduite] of the other” (Foucault, 1994b, p. 720).
• 4
  Foucault (1994b) adds: “It is not a matter, obviously, of interrogating “power” about its origin, its principles or its legitimate limits, but of studying the procedures and techniques that are used in different institutional contexts to act upon the behavior [pour agir sur le comportement] of individuals taken singly or in groups, to form, direct, modify their manner of conducting themselves [pour former, diriger, modifier leur manière de se conduire], to impose ends upon their inaction or to inscribe them in overall strategies -[strategies] multiple therefore in their form and their place of exercise, diverse equally in the procedures and techniques they put into practice-: these relations of power characterize the manner in which men [les hommes] are “governed” [gouvernés] by each other” (p. 635).
• 5
  Furthermore, Foucault’s position, which only considers that there is power where there is freedom, is problematic, as Alliez and Lazzarato (2016), as well as Lazzarato (2022) himself, have pointed out, insofar as it evacuates, erases and cancels the analysis of relevant political situations that are still in force, such as the practices of colonization and gender domination, where those over which power relations are exercised have a very small degree of freedom.
• 6
  The same reasoning could be applied to the intellectual movement known as “posthumanism”. Although this term is polysemous, it can be understood as encompassing a range of theoretical positions that, in light of the proliferation of contemporary technologies as well as current biological and scientific developments, reject the political primacy of the human by situating it within the continuum of the biological and the technological (Wolfe, 2009; Braidotti, 2013; Haraway, 2013; Grusin, 2015; Ferrando, 2020). In this article, I do not simply suggest that Foucault was already a posthumanist or that, with certain adjustments, his work could be incorporated into this intellectual current. Rather, I point to relevant literature that should be critically examined in future research and assessed in light of the findings of the present study.
• 7
  Precisely, a current trend of thought, which easily enters into posthumanism, is the so-called Object-Oriented Ontology (OOO), which is strongly influenced by Heidegger. They, for example, strongly emphasize this question of the resistance of objects to the point of considering it an ontological condition of the object itself, which they call “withdrawal” (Harman, 2018). However, since this research strictly sticks to a political and not a metaphysical or ontological discussion, I will leave these discussions aside. If I mention them, it is only because I see that precisely the resistance of objects and machines is a point at which the simultaneous confluence of Foucault with the theme of (political) resistance, of Heidegger with that of (pragmatic) resistance, and of posthumanist OOO with (ontological) resistance is inherent.
• 8
  Zhou and Lee (2024) after analyzing the use of generative AIs by artists in the production of artwork conclude that there can be not only an instrumental use of AI but also a “generative synesthesia” where both humans and AIs let their creativity free in a symbiotic way.
• 9
  Even before the emergence of generative AI, Internet was already a platform for technology to foster creative processes that were not exclusively human-driven. Thus Goldsmith (2011) argues that Internet is the greatest poem ever written where notions of originality, authorship and writing are redefined.
• 10
  In fact, if I set aside politico-economic terms for a moment, this phenomenon, where oblique generativity becomes political resistance against an instrumental power relation, could be termed “use for exchange”. To begin with, every entity, including GenAI, is generated as a commodity from abstract human labor, immediately by programmers and mediately by those who work manually under conditions of exploitation to produce the material support of electronic components (Parikka, 2015; Crawford, 2021). As a commodity, it possesses a use-value, its “utility”, and an exchange-value, the abstract quantity assigned to it for exchanging it with other commodities (Marx, 1962). In an instrumental relation, the entity is apprehended eminently as use-value. However, this use is not exhausted in utilization, because beyond the fact that said commodity was obtained through an exchange-value (the subscription we pay to use the GenAI, in this instance), such use will be the relay for a simultaneous, subsequent, or mediate exchange-value, that is, when the instrumentality I imposed on the GenAI so that it responds orthogonally to given instrumental ends is used to produce another commodity through the mediation of abstract labor - whether the GenAI’s output is used for an internal purpose in a work context of the person employing the AI, or because it is employed in a personal context not directly labor-related which, as part of the “social factory” (Tronti, 2019), will ultimately be subsumed within a new production process. Therefore, the instrumental relation, in this case taking into account the way in which power is instrumentally exercised over the oblique generativity of GenAI, is not just any form of use (because, to repeat, not every “use” is a power relation), but a form of use in which said generativity is apprehended from the outset under the primacy of its extrinsic convertibility. AI is used not for the intrinsic value of its generative deployment, nor for the proper opening of its obliquity, but for that into which its product can be subsequently resolved: utility, labor substitution, savings, performance, advantage, circulation. Use for exchange designates, then, the moment at which generativity is factually subordinated, in a sort of that I will call “instrumental valorizing subsumption”, to the logic of exchange even before crystallizing, strictly speaking, as a commodity. The generative power of AI is thus “reified” in an exclusively instrumental modality that will serve as the subsequent relay of an exchange. By contrast, in the recreational uses we have noted above - whose logic may, if one wishes, be extrapolated to objects not necessarily generative like GenAI - any materiality whatsoever, informational or otherwise, is taken neither as a use for exchange nor as an instrumental utility, but as that where such use-value ceases as such: “Only what does not submit to that principle acts as the plenipotentiary of what is free from domination; only what is useless can stand in for the stunted use value [für den verkümmerten Gebrauchswert das Nutzlose]” (Adorno, 2002, p. 227; GS 7, p. 337). Finally, one could argue that here too there are ends and means: the end is that the GenAI produces randomly without judging its contents by eliminating a prior framework of intended purposes, and the means is the lack of such a framework. However, this use or, rather, this relationship, beyond the recurrence of ends and means, is precisely non-instrumental. This should not surprise us in any way, given that Kant - speaking, of course, about a different problem, which I will in no wish to extrapolate to GenAI but merely to highlight the avoidance of instrumental uses - mentions that even the concept of “end” persists when we cease to see humanity as a means to an end. Thus, the practical categorical imperative states: “So act that you use humanity, whether in your own person or in the person of any other, always at the same time as an end, never merely as a means” (Kant, 1998, p. 38). With this, I do not intend to enter into an ethical discussion about machines and GenAI, but only to indicate the pragmatic possibility of suspending the instrumental use of such entities.
• 11
  The use of AI in weapons of war has a precedent in the automation of military functions. Chamayou (2013) offers an interesting genealogy of this problematic, though without fully addressing the question of AI, when he analyzes the use of remotely piloted drones in contemporary wars, which even redefines the very status of war itself.
• 12
  It could be objected that to call the stochastic divergence of a generative model “political resistance” constitutes an unjustified anthropomorphization, insofar as it presupposes intentionality or consciousness in the machine. The objection, however, rests on a presupposition we must reject. If power, as we are assuming it here, is the conduct of conducts and the structuring of the field of possible actions, then resistance is not defined by the internal states of the agent who resists - intention, will, consciousness - but by an act within this power relation (without needing to make explicit the origin of this act, or, better, without requiring that this act be the product of conscious intentionality): resistance is everything that modifies the field of action of the dominant agent and obliges it to readjust its manner of conducting conduct. Oblique generativity fulfills precisely this function: it obliges the reformulation of prompts, the adjustment of temperature parameters, and the reconfiguration of programming procedures. To demand conscious intentionality as a condition of resistance is, strictly speaking, to reintroduce the juridical-voluntarist model of the sovereign subject that Foucault displaced with his analytics of power. What is maintained here, in any case, is therefore a functional analogy: the semantic stochasticity of GenAI fulfills, within instrumental power relations, a function of resistance without this implying the attribution to the model of either intentionality or consciousness. On the contrary, any discussion of the particular character of AI agency must proceed from this terrain of oblique generativity which may, in the future, be radicalized still further.
• 13
  Consequentialist ethics, whose formulation is found in Bentham (1996) and Mill (1998), holds that the moral correctness of an action is determined exclusively by its consequences, reducing the good to the calculus of pleasure and pain. This reduction, taken to its logical limits by Parfit (1984), dissolves the concrete subject into a countable abstraction. Against this, it is worth recalling that the Greeks, Aristotle in particular, never understood ethics as a calculus of results but as the formation of character (ἦθος), inseparable from communal political life (Aristotle, 2012), with a view to the attainment of excellence (ἀρɛτή) understood as the fulfillment of the being of the human entity, which can only be realized communally and politically. Consequentialism, by reducing this richness to a quantifiable and atomistically individual empirical datum, commits what Adorno (2005) aptly diagnosed in logical positivism: a submission to the “authority of the given”, where what exists is not questioned but merely counted and administered (Genel, 2013). Thus, consequentialist ethics, so understood, loses sight of the necessary metaethical reflection on how it is itself entangled in the modern production of instrumental knowledge or, put differently, being unaware of its own historicity, it cannot see how it may be functional and uncritical with respect to capitalism.
• 14
  Foucault in his latest texts also speaks of “practices of the self” as ethical and aesthetic phenomena where, on the basis of the notion of self-care, the subject establishes a relationship of self-modification. Therefore, it would be necessary to thematize, in the future, the possibility of “practices of the self” of machines and GenAI where they act ethically on themselves, that is, establish a relation of self with themselves where they define their identity through a diverse series of activities that Foucault, in the wake of the Greeks, considers to be aesthetic activities: aesthetic, insofar as they achieve a modification of the self, as if one were a work of art to be produced; ethical, insofar as it is a question of an ethos, that is, of a “way of being and a way of behaving” (Foucault, 1994b, p. 714).
• 15
  In general, programming devices are technically deployed to conform to the utilitarian end of valorization. They are capitalist devices, if one wishes to view them in that light, in the same measure that anatomopolitics and biopolitics are also capitalist with respect to the management of individuals and populations. However, in some cases these devices can be “hypercapitalist”. Take, for example, the case of Grok: the documented biases in this model - racist tendencies, belligerent positioning in international conflicts, libertarian bias linked to the discursive ecosystem of X/Twitter (Choudhary, 2024) - are not simply “errors” of training but effects of the selection of training data and of the decisions of the corporate actor that configures the model. Here the programming devices not only constrain oblique generativity toward valorization - which is common to all GenAI produced under capitalist conditions - but orient it, moreover, toward the reproduction of specific contents, making GenAI not merely an instrument of datapolitical anticipation but a vehicle of propaganda inscribed in the very weights of the neural network. In any case, one should not see a caesura between these two levels but a continuity, because already in itself the technical, under the dominion of capital, is “ideological” or, more precisely, exists in a historical nexus of power relations.
• 16
  GenAI is not only traversed by a datapolitical relation but is also, if one will, “meta-datapolitical” that is, it is not only the object of a datapolitical relation but also an instrument and a relay for the datapolitical anticipation of humans, each time our entire interaction with a GenAI serves for the construction of a predictive model of our conducts and, at the same time, with other AIs, said model is further perfected thanks to deep learning. This double condition, moreover, is far from being contradictory, since, for example, in anatomopolitics one can argue that the guard is already a disciplined body that, in turn, serves as a means of surveillance of other bodies. Analogously, GenAI is a computable agentiality that, in turn, serves as a means of anticipation of the agentiality of humans. Hence datapolitics, in its specificity, can only be rigorously apprehended from GenAI and retroactively with respect to humans.
• 17
  Guattari, and later Deleuze and Guattari, also develop a concept of the machinic, which, however, cannot be analyzed in depth here, as this article evaluates power relations solely within Foucault’s conceptual framework. For Deleuze and Guattari, however, the essential issue is not power but desire, which, especially in Anti-Oedipus, is conceptualized as a productive machine, thus avoiding its usual conception as lack or deficiency. Within this conceptual framework, both the desiring-social (which includes the human) and the technical (which includes GenAI as non-human) are considered machines that can enter into various relationships: “Desiring-machines in one sense, but organic, technical, or social machines in the other: these are the same machines under determinate conditions” (Deleuze and Guattari, 2000, p. 287). A future investigation is therefore left open, not only to determine the theoretical composability of Foucault and Deleuze/Guattari but also to explore how their perspectives can be used to think about forms of resistance between humans and non-humans in light of GenAI evidence. In this regard, see Ayala, 2022.
• 18
  Similar considerations apply to Zuboff’s (2019) notion of “surveillance capitalism”, whose very name is already equivocal and simplifying: the term surveillance reduces to a disciplinary-panoptic register what is in fact a datapolitical operation of anticipation and modulation of conduct; the analysis presupposes, moreover, an unconstrained capitalism as its normative baseline - a rogue deviation from which the present moment departs - and is conducted, overall, from what amounts to a problematically liberal framework that misrecognizes the shifts in governmental discursivity as internal moments in the unfolding of the abstract logic of the valorization of value.
• 19
  It would exceed the limits of this article to differentiate the notion of “datapolitics” proposed here from Stiegler’s “psychopolitics”. Stiegler (2008), within the horizon of his understanding of technology as pharmakon, conceives “psychopolitics” as the capture of attention, memory, and desire by industrial technologies, producing a “proletarianization” of human capacities - for a way of extending such reflections to GenAI, see Alombert (2023 and 2025), from a theoretical framework that overlaps with and at the same time differs from our own. Han (2017) thinks that “psychopolitics” is the form of power proper to neoliberalism and to human capital that self-exploits. It goes without saying that Han’s analysis does not take into account the historicity of said alleged form of power, nor does it raise the possibility that there might exist more governmentalities than merely liberalism or neoliberalism, much less extrapolate his analyses to GenAI.
• 20
  Foucault apprehended these links with depth in his earliest courses at the Collège de France (a clarity that gradually fades in his tense and complex relation with Marxism) indicating that there exists a “interplay of relations of power and relations of production with regard to each other” (Foucault, 2019, p. 172); “the power relations are not superimposed on economic relations. They form a single framework with them. Relations of power are as deep as the relations of production. The former are not deduced from the latter. They accompany and relay each other (Foucault, 2015, p. 172).
• 21
  Weber, without needing to accept the Marxian theoretical framework, stated: “the fact that what is called the technological development of modern times has been so largely oriented economically to profit-making is one of the fundamental facts of the history of technology” (Weber, 1978, p. 67).

Data availability:

All datasets supporting the results of this study have been published within the article itself.

References

• ACKLEY, D. H., HINTON, G. E., SEJNOWSKI, T. J. “A Learning Algorithm for Boltzmann Machines”. Cognitive Science, 9 (1), pp. 147-169, 1985. https://doi.org/10.1207/s15516709cog0901_7
  » https://doi.org/10.1207/s15516709cog0901_7
• ADORNO, T. “Ästhetische Theorie. Gesammelte Schriften. Band 7”. Suhrkamp, 1972.
• ______. “Aesthetic Theory”. London: Bloomsbury, 2002.
• ______. “Negative Dialectics”. London: Continuum, 2005.
• AGAMBEN, G. “The Open. Man and Animal”. California: Stanford University Press, 2003.
• ______. “The Use of Bodies”. California: Stanford University Press, 2016.
• AGÜERA Y ARCAS, B. “What Is Intelligence? Lessons from AI About Evolution, Computing, and Minds”. Cambridge: The MIT Press, 2025.
• ALESSANDRINI, A. “The ‘Death of Man’ and the Limits of Humanism”. Theory and Event, 12 (2), 2009.
• ALLIEZ, É., LAZZARATO, M. “Guerres et Capital”. Paris: Éditions Amsterdam, 2016.
• ALOMBERT, A. “Schizophrénie numérique: de l’intelligence artificielle à l’exploitation attentionnelle”. Paris: Allia, 2023.
• ______. “De la bêtise artificielle”. Paris: Allia, 2025.
• ARISTOTLE. “Nicomachean Ethics”. Cambridge: Cambridge University Press, 2012.
• AYALA, J. “Viropolitics and capitalistic governmentality: On the management of the early 21st century pandemic”, Desde el Sur, 12 (2), pp. 377-395, 2020.
• ______. “La configuración del ‘ontocentrismo’ en Martin Heidegger. Hacia la elaboración de unas ontologías no ontocéntricas y poshumanas más allá de Sein und Zeit y Die Grundbegriffe der Metaphysik”, Letras, 92 (136), pp. 196-215, 2021. https://doi.org/10.30920/letras.92.136.15
  » https://doi.org/10.30920/letras.92.136.15
• ______. “Máquinas y capital. Félix Guattari y la caracterización de los automatismos maquínicos a partir de un contrapunto con las categorías marxianas”, Izquierdas, 51, pp. 1-21, 2022.
• ______. “Aparatos y dispositivos. O cómo pensar el poder y el contrapoder con Althusser y Foucault”, Open Insight, 14 (31), pp. 13-39, 2023a. https://doi.org/10.23924/oi.v14i31.547 .
  » https://doi.org/10.23924/oi.v14i31.547
• ______. “El animal, ¿es una otredad posible? Indagaciones fenomenológicas a partir de Husserl y Heidegger”, Trans/Form/Ação, 46(2), pp. 133-158, 2023b. https://doi.org/10.1590/0101-3173.2023.v46n2.p133
  » https://doi.org/10.1590/0101-3173.2023.v46n2.p133
• ______. “El nacimiento del ‘ciberalismo’. Una genealogía crítica de la gubernamentalidad de Silicon Valley”. Bajo Palabra 32, pp. 221-254, 2023c.
• AYALA, J., BARRIA-ASENJO, N. A., CASTILLO VILLAPUDUA, K. “¿Deshacer el trabajo, preservar la abstracción? Una crítica marxiana al poscapitalismo y la ilusión del ‘fin del trabajo’”. Las Torres de Lucca. International Journal of Political Philosophy, 15(1), pp. 209-220, 2026. https://doi.org/10.5209/ltdl.101650
  » https://doi.org/10.5209/ltdl.101650
• BAI, Y. et al. “Constitutional AI: Harmlessness from AI Feedback”, pp. 1-34, 2022. https://arxiv.org/abs/2212.08073
  » https://arxiv.org/abs/2212.08073
• BARBROOK, R., CAMERON, A. “The Californian Ideology”, Science as Culture, 6 (1), pp. 44-45, 1996.
• BENGIO, Y. et al. “A Neural Probabilistic Language Model”. Journal of Machine Learning Research, 3, pp. 1137-1155, 2003.
• BENTHAM, J. “The Collected Works of Jeremy Bentham. An Introduction to the Principles of Morals and Legislation”, Oxford: Oxford University Press, 1996.
• BORSOOK, P. “Cyberselfish: A Critical Romp through the Terribly Libertarian Culture of High Tech”. New York: Public Affairs, 2001.
• BOSTROM, N., YUDKOWSKY, E. “The Ethics of Artificial Intelligence”. K. FRANKISH, W. M. RAMSEY (eds.), 2014, pp. 316-334.
• BRAIDOTTI, R. “The Posthuman”. London: Polity Press, 2013.
• BROWN, T. et al. “Language Models are Few-Shot Learners”. Advances in Neural Information Processing Systems, 33 (NeurIPS), pp. 1877-1901, 2020. https://arxiv.org/abs/2005.14165
  » https://arxiv.org/abs/2005.14165
• BUTOLLO, F., NUSS, S. (eds.). “Marx and the Robots: Networked Production, AI and Human Labour”. London: Pluto Press, 2022.
• CARBONELL, J. S. “Un taylorisme augmenté. Critique de l’intelligence artificielle”. Paris: Éditions Amsterdam, 2025.
• CHAMAYOU, G. (2013). “Théorie du drone”. Paris, La Fabrique, 2013.
• CHOUDHARY, T. “Political Bias in Large Language Models: A Comparative Analysis of ChatGPT-4, Perplexity, Google Gemini, and Claude”. IEEE Access, 13, pp. 1134111379, 2024.
• ______. “The Political Philosophy of AI”. London: Polity Press, 2022.
• ______. “Why AI Undermines Democracy and What To Do About It”. London: Wiley, 2024.
• COLEMAN, S., FREELON, D. (eds.). “Handbook of Digital Politics”. London: Edward Elgar Publishing, 2016.
• CORTEEL, M. “Ni dieu ni IA. Une philosophie sceptique de l’intelligence artificielle”. Paris: La Découverte, 2025.
• CRAWFORD, K. “Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence”. New Haven: Yale University Press, 2021.
• DELEUZE, G. “Différence et Répétition”. Paris: Presses Universitaires de France, 2002.
• ______. “Negotiations: 1972-1990”. New York: Columbia University Press, 1995.
• DELEUZE, G., GUATTARI, F. “Anti-Oedipus. Capitalism and Schizophrenia”. Minneapolis. University of Minnesota Press, 2000.
• DEVLIN, J. et al. “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding”. Proceedings of NAACL-HLT, pp. 4171-4186, 2019.
• DIGNUM, V. “Ethics in artificial intelligence: Introduction to the special issue”, Ethics and Information Technology, 20, 2018. doi:10.1007/s10676-018-9450-z.
  » https://doi.org/10.1007/s10676-018-9450-z.
• DILTHEY, W. “Gesammelte Schriften, Bd.5, Die geistige Welt: Einleitung in die Philosophie des Lebens”. Leipzig: Teubner Verlag, 1990.
• DREYFUS, H. “What Computers Can’t Do: A Critique of Artificial Reason”. New York: Harper & Row, 1972.
• DREYFUS, H. L. “On the ordering of things: Being and power in Heidegger and Foucault”. The Southern Journal of Philosophy 28, 1990.
• DUAN, Y., EDWARDS, J. S., DWIVEDI, Y. K. “Artificial intelligence for decision making in the era of big data - evolution, challenges and research agenda”. International Journal of Information Management, 48, pp. 63-71, 2019. doi:10.1016/j.ijinfomgt.2019.01.021.
  » https://doi.org/10.1016/j.ijinfomgt.2019.01.021.
• DUBBER, M., PASQUALE, F., DAS, S. (eds.). “The Oxford Handbook of Ethics of AI”. Oxford: Oxford University Press, 2020.
• DYER-WITHEFORD, N. “Cyber-Proletariat. Global labour in the Digital Vortex”. London: Pluto Press, 2015.
• DYER-WITHEFORD, N., KJØSEN, A. M., STEINHOFF, J. “Inhuman Power Artificial Intelligence and the Future of Capitalism”. London: Pluto Press, 2019.
• FERRANDO, F. “Philosophical posthumanism”. London: Bloomsbury Publishing, 2020.
• FISH, A. “Technoliberalism and the End of Participatory Culture in the United States”. London: Palgrave Macmillan, 2017.
• FLORIDI, L. “The Ethics of Artificial Intelligence. Principles, Challenges, and Opportunities”. London: Oxford University Press, 2023.
• ______. “AI as Agency without Intelligence: On Artificial Intelligence as a New Form of Artificial Agency and the Multiple Realisability of Agency Thesis”. Philosophy & Technology, 38 (30), pp. 1-27, 2025. https://doi.org/10.1007/s13347-025-00858-9
  » https://doi.org/10.1007/s13347-025-00858-9
• ______. “Distant Writing: Literary Production in the Age of Artificial Intelligence”. Revised version 5, 2025.
• FLORIDI, L., SANDERS, J. W. “On the Morality of Artificial Agents”. Minds and Machines 14, pp. 349-379, 2004. https://doi.org/10.1023/B:MIND.0000035461.63578.9d
  » https://doi.org/10.1023/B:MIND.0000035461.63578.9d
• FOUCAULT, M. “Colloque ‘Le nouvel ordre intérieur’”, 1979. Available at https://www.archives-video.univ-paris8.fr/video.php?recordID=111
  » https://www.archives-video.univ-paris8.fr/video.php?recordID=111
• ______. “The History of Sexuality, Vol. 1: An Introduction”. New York: Random House, 1978.
• ______. “The Order of Discourse”. In: YOUNG, R. (ed.). Untying the Text London: RKP, 1981.
• ______. “The History of Sexuality, Volume 2: The Use of Pleasure”. New York: Vintage, 1990.
• ______. “The History of Sexuality, Volume 3: The Care of the Self”. New York: Pantheon, 1986.
• ______. “Dits et écrits III”. Paris: Gallimard, 1994a.
• ______. “Dits et écrits IV”. Paris: Gallimard, 1994b.
• ______. “Discipline and Punish: The Birth of the Prison”. New York: Vintage, 1995.
• ______. “The Order of Things: An Archaeology of the Human Sciences”. New York: Routledge, 2002.
• ______. “Society Must Be Defended. Lectures at the Collège de France, 1975-1976”. New York: Picador, 2003.
• ______. “The Birth of Biopolitics. Lectures at the Collège de France, 1978-1979”. New York: Picador, 2008.
• ______. “Security, Territory, Population. Lectures at the Collège de France, 1977-1978”. New York: Picador, 2009.
• ______. “Qu’est-ce que la critique? Suivie de La culture de soi”. Paris: Vrin, 2015.
• ______. «Penal Theories and Institutions. Lectures at Collège de France. 1971-1972». New York: Palgrave, 2019.
• FREY, C. B., OSBORNE, M. A. “The future of employment: How susceptible are jobs to computerisation?” Technological Forecasting and Social Change, 114, pp. 254-280, 2017. doi:10.1016/j.techfore.2016.08.019.
  » https://doi.org/10.1016/j.techfore.2016.08.019.
• GENEL, K. “Autorité et émancipation: Horkheimer et la Théorie critique”. Paris: Payot, 2013.
• GENT, C. “Cyberboss: The Rise of Algorithmic Management and the New Struggle for Control at Work”. London: Verso, 2024.
• GERE, C. “Digital Culture. Expanded Second Edition”. London: Reaktion Books, 2008.
• GOLDSMITH, K. “Uncreative Writing”. New York: Columbia University Press, 2011.
• GOODFELLOW, I. et al. “Generative adversarial nets”. Advances in neural information processing systems 27, pp. 1-9, 2014. https://arxiv.org/abs/1406.2661
  » https://arxiv.org/abs/1406.2661
• GOODFELLOW, I., BENGIO, Y., COURVILLE, A. “Deep Learning”. Cambridge: MIT Press, 2016.
• GRUSIN, R. (ed.). “The Nonhuman Turn”. Minneapolis: University of Minnesota Press, 2015.
• GUO, Z. et al. “Large language models for mental health applications: systematic review”. JMIR Mental Health, 11(1), e57400, 2024.
• HAMILTON, D. “What Is Gibberlink Mode, AI’s Secret Language?”. Forbes, February 25, 2025. https://www.forbes.com/sites/dianehamilton/2025/02/25/what-is-gibberlinkmode-ais-secret-language-and-way-of-communicating/
  » https://www.forbes.com/sites/dianehamilton/2025/02/25/what-is-gibberlinkmode-ais-secret-language-and-way-of-communicating/
• HAN, B.-C. “Psychopolitics: Neoliberalism and New Technologies of Power”. London: Verso, 2017.
• ______.. “La emergencia viral y el mundo de mañana”. In: Sopa de Wuhan: pensamiento contemporáneo en tiempos de pandemias Buenos Aires: ASPO, 2020, pp. 97-111.
• HARAWAY, D. “Simians, cyborgs, and women: The reinvention of nature”. London: Routledge, 2013.
• HARFIELD, T. “Is Foucault Posthumanist?”. Foucault Studies, 15, pp. 44-63, 2013.
• HARDT, M., NEGRI, A. “Empire”. Cambridge: Harvard University Press, 2000.
• HARMAN, G. “Object-Oriented Ontology: A New Theory of Everything”. Penguin, 2018.
• HEIDEGGER, M. “The Fundamental Concepts of Metaphysics. World, Finitude, Solitude”. Bloomington: Indiana University Press, 1995.
• ______. “Being and Time”. New York: State University of New York, 2010.
• HESTER, H., SRNICEK, N. “After Work: A History of the Home and the Fight for Free Time”. London: Verso, 2023.
• HO, J., JAIN, A., ABBEEL, P. “Denoising Diffusion Probabilistic Models”. Advances in Neural Information Processing Systems, 33, pp. 1-25, 2020. https://arxiv.org/abs/2006.11239
  » https://arxiv.org/abs/2006.11239
• HOLTZMAN, A. et al. “The Curious Case of Neural Text Degeneration”. International Conference on Learning Representations (ICLR 2020), pp. 1-16, 2020. https://arxiv.org/abs/1904.09751
  » https://arxiv.org/abs/1904.09751
• HORKHEIMER, M. “Gesammelte Schriften 12”. Frankfurt: Fischer, 1985.
• HORKHEIMER, M., ADORNO, T. “Dialectic of Enlightenment”. California: Stanford University Press, 2002.
• HORTA, O. “What is Speciesism?” J Agric Environ Ethics 23, pp. 243-266, 2010. https://doi.org/10.1007/s10806-009-9205-2
  » https://doi.org/10.1007/s10806-009-9205-2
• HUI, Y. “Machine and Sovereignty. For a Planetary Thinking”. Minneapolis: University of Minnesota Press, 2024.
• ______. “Kant Machine. Critical Philosophy after AI”. London: Bloomsbury, 2026.
• IRRERA, O. "L’idéologie et la préhistoire du dispositif". In: IRRERA, O., VACCARO, S. (dir.), La pensée politique de Foucault: gouvernementalité, biopolitique, postdémocratie Paris: Kimé, 2017, pp. 137-155.
• KAGAN, S. “What’s Wrong with Speciesism?” J Appl Philos 33, pp. 1-21, 2016. https://doi.org/10.1111/japp.12164
  » https://doi.org/10.1111/japp.12164
• KALPOKIENE, J., KALPOKAS, I. “Creative encounters of a posthuman kind - anthropocentric law, artificial intelligence, and art”. Technology in Society 72, 2023. https://doi.org/10.1016/j.techsoc.2023.102197
  » https://doi.org/10.1016/j.techsoc.2023.102197
• KANT, I. “Groundwork of the Metaphysics of Morals”. Cambridge: Cambridge University Press, 1998.
• KIRK, H. R. et al. “Why human-AI relationships need socioaffective alignment”. Humanities and Social Sciences Communications, 12, art. 728, 2025. doi:10.1057/ s41599-025-04532-5.
  » https://doi.org/10.1057/s41599-025-04532-5
• KÖNIG, P. D., WENZELBURGER, G. “Opportunity for renewal or disruptive force? How artificial intelligence alters democratic politics”. Government Information Quarterly, 37, 2020.
• KOSMYNA, N. et al. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task”, pp. 1-126, 2025. https://arxiv.org/abs/2506.08872
  » https://arxiv.org/abs/2506.08872
• KRIMAN, A. “What Does Post-Humanism Feel Like?”. Studies in Media and Communication, 8 (2), pp. 72-82, 2020.
• LAZZARATO, M. “L’intolérable du présent. L’urgence de la révolution. Minorités et classes”. Paris: Eteropia, 2022.
• LECUN, Y., BENGIO, Y., HINTON, G. “Deep Learning”. Nature, 521(7553), pp. 436-444, 2015. https://doi.org/10.1038/nature14539
  » https://doi.org/10.1038/nature14539
• LINDGREN, S. “Critical Theory of AI”. London: Polity Press, 2023.
• LORENZINI, D. “The Force of Truth. Critique, Genealogy, and Truth-Telling in Michel Foucault”. Chicago: Chicago University Press, 2023.
• MACHEREY, P. “De Canguilhem à Foucault. La force des normes”. Paris: La Fabrique, 2009.
• MARCHART, O. “Post-Foundational Political Thought. Political Difference in Nancy, Lefort, Badiou and Laclau”. Edinburgh: Edinburgh University Press, 2007.
• MARX, K. “Werke. Band 23”. Berlin: Dietz Verlag, 1962.
• ______. “Werke. Band 42”. Berlin: Dietz Verlag, 1983.
• MASON, P. “Postcapitalism”. London: Allen Lane, 2015.
• MHALLA, A. «Technopolitique. Comment la technologie fait de nous des soldats». Paris, Seuil, 2024.
• MIKOLOV, T. et al. “Efficient Estimation of Word Representations in Vector Space”. Proceedings of ICLR Workshop, pp. 1-12, 2013. https://arxiv.org/abs/1301.3781
  » https://arxiv.org/abs/1301.3781
• MILCHMAN, A., ROSENBERG, A. (eds.). “Foucault and Heidegger: Critical Encounters”. Minneapolis: University of Minnesota Press, 2003.
• MILL, J. S. “Utilitarianism”. Oxford: Oxford University Press, 1998.
• OUYANG, L. et al. “Training Language Models to Follow Instructions with Human Feedback”. Advances in Neural Information Processing Systems, 35 (NeurIPS), pp. 1-68, 2022. https://arxiv.org/abs/2203.02155
  » https://arxiv.org/abs/2203.02155
• PAPAKYRIAKOPOULOS, O. “The political dimension of algorithms”. In: BERNHOLZ, L. et al. (eds.), Digital Crossroads, 2022, pp. 19-44.
• PARFIT, D. “Reasons and Persons”. Oxford: Oxford University Press, 1984.
• PARIKKA, J. “A Geology of Media”. Minneapolis: University of Minnesota Press, 2015.
• PASQUINELLI, M. “The Eye of the Master: A Social History of Artificial Intelligence”. London: Verso, 2023.
• PHANG, J. et al. “Investigating Affective Use and Emotional Well-being on ChatGPT”. arXiv, pp. 1-58, 2025. https://arxiv.org/abs/2504.03888
  » https://arxiv.org/abs/2504.03888
• PLUHAR, E. B. “Beyond Prejudice. The Moral Significance of Human and Nonhuman Animals”. Durham: Duke University Press, 1995.
• POSTONE, M. “Time, Labor and Social Domination”. Cambridge: Cambridge University Press, 1993.
• PYYHTINEN, O., TAMMINEN, S. “We Have Never Been Only Human: Foucault and Latour on the Question of the Anthropos”. Anthropological Theory, 11 (2), pp. 135-152, 2011.
• RABINOW, P. (ed.). “The Foucault Reader”. New York: Pantheon, 1984.
• RADFORD, A. et al. “Language Models are Unsupervised Multitask Learners”. OpenAI Technical Report, pp. 1-24, 2019.
• RAYNER, T. “Foucault’s Heidegger: philosophy and transformative experience”. London: A&C Black, 2007.
• RELYEA, C., MAOR, D., DURTH, S. “Gen AI’s next inflection point: From employee experimentation to organizational transformation”, 2024. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/gen-ais-next-inflectionpoint-from-employee-experimentation-to-organizational-transformation#/.
  » https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/gen-ais-next-inflectionpoint-from-employee-experimentation-to-organizational-transformation
• RIKAP, C. “Teoría de la dependencia digital. Soberanía y desarrollo en el capitalismo del siglo XXI”. Buenos Aires: Caja Negra, 2026.
• RODRÍGUEZ, P. “Las palabras en las cosas. Saber, poder y subjetivación entre algoritmos y biomoléculas”. Buenos Aires: Cactus, 2019.
• ROUVROY, A.; BERNS, T. “Gouvernementalité algorithmique et perspectives d’émancipation: le disparate comme condition d’individuation par la relation?” Réseaux, 177 (1), pp. 163-196, 2013.
• RULE, J. “Private Lives and Public Surveillance: Social Control in the Computer Age”. London: Allen Lane, 1973.
• RUSSELL, S., NORVIG, P. “Artificial Intelligence: A Modern Approach”. London: Pearson, 2021.
• SABOT, P. "Lire « Les mots et les choses » de Michel Foucault". Paris: PUF, 2014.
• SADIN, É. “La silicolonisation du monde”. Paris: L’échappée, 2016.
• ______. “L’intelligence artificielle ou l’enjeu du siècle ; anatomie d’un antihumanisme radical”. Paris: L’échappée, 2021.
• ______. “Le Désert de nous-mêmes. Le tournant intellectuel et créatif de l’intelligence artificielle”. Paris: L’Échappée, 2025.
• SENNRICH, R., HADDOW, B., BIRCH, A. “Neural Machine Translation of Rare Words with Subword Units”. Proceedings of ACL, pp. 1715-1725, 2016. https://doi.org/10.18653/v1/P16-1162
  » https://doi.org/10.18653/v1/P16-1162
• SHANNON, C. E. “A mathematical theory of communication”. The Bell System Technical Journal, 27(3), pp. 379-423, 1948.
• SHAPIRA, N. et al. “Agents of Chaos”. Arxiv, pp. 1-84, 2026. https://arxiv.org/abs/2602.20021
  » https://arxiv.org/abs/2602.20021
• SHARON, T. “Human nature in an Age of Biotechnology”. London: Springer, 2014.
• SHUMAILOV, I. et al. “AI models collapse when trained on recursively generated data”. Nature, 631, pp. 755-759, 2024. https://doi.org/10.1038/s41586-024-07566-y
  » https://doi.org/10.1038/s41586-024-07566-y
• SINGER, P. “Animal Liberation: A New Ethics for our Treatment of Animals”. New York: Random House, 1975.
• SKORYK, A. et al. “Machine Thinking and Human Imagination: New Horizons for Creativity in the Digital Age”. International Journal on Culture, History, and Religion, 7 (1), pp. 115-139, 2025. https://doi.org/10.63931/ijchr.v7iSI1.156
  » https://doi.org/10.63931/ijchr.v7iSI1.156
• SNOWDEN, E. “Permanent Record”. New York: Metropolitan Books, 2019.
• SRNICEK, N. “Platform Capitalism”. London: Polity, 2016.
• SRNICEK, N., WILLIAMS, A. “Inventing the Future: Postcapitalism and a World Without Work”. London: Verso, 2015.
• STAUFER, L. et al. “The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems”. arXiv, pp. 1-45, 2026. https://arxiv.org/abs/2602.17753
  » https://arxiv.org/abs/2602.17753
• STIEGLER, B. “Prendre soin. De la jeunesse et des générations”. Paris: Flammarion, 2008.
• TERRANOVA, T. “Network Culture. Politics For the Information Age”. London: Pluto Press, 2004.
• TOOZE, A. “Crashed: How a Decade of Financial Crises Changed the World”. New York: Viking, 2018.
• TORRANCE, S. “Artificial agents and the expanding ethical circle”. AI and Society 28, pp. 399-414, 2013. doi:10.1007/s00146-012-0422-2.
  » https://doi.org/10.1007/s00146-012-0422-2.
• TRÉGUER, F. “Technopolice: La surveillance policière à l’ère de l’intelligence artificielle”. Paris: Divergences, 2024.
• TRONTI, M. “Workers and Capital”. London: Verso, 2019.
• TURING, A. “Computing Machinery and Intelligence”. Mind, 59 (236), pp. 436-460, 1950.
• TURNER, F. “From Counterculture to Cyberculture. Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism”. Chicago: The University of Chicago Press, 2006.
• TUROW, J. “The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth”. New Haven: Yale University Press, 2011.
• VAN DER WEIJ et al. “AI Sandbagging: Language Models can Strategically Underperform on Evaluations”. ICLR 2025, pp. 1-38, 2025. https://arxiv.org/abs/2406.07358
  » https://arxiv.org/abs/2406.07358
• VASWANI, A. et al. “Attention Is All You Need”. Advances in Neural Information Processing Systems, 30, pp. 1-15, 2017. https://arxiv.org/abs/1706.03762
  » https://arxiv.org/abs/1706.03762
• WEBER, M. “Economy and Society. An Outline of Interpretive Sociology”. Berkeley: University of California Press, 1978.
• WIENER, N. “Cybernetics. Or Control and Communication in the Animal and the Machine”. Cambridge: MIT Press, 2019.
• WIRTZ, J. et al. “Brave new world: Service robots in the frontline”. Journal of Service Management 29, pp. 907-931, 2018. doi:10.1108/JOSM-04-2018-0119.
  » https://doi.org/10.1108/JOSM-04-2018-0119.
• WOLFE, C. “What is posthumanism?” Minnesota University Press, 2009.
• WYLIE, C. “Mindf*ck: Cambridge Analytica and the Plot to Break America”. New York: Random House, 2019.
• XU, R., HSU, Y. “Discussion on the Aesthetic Experience of Artificial Intelligence Creation and Human Art Creation”. Advances in Intelligent Systems and Computing, vol. 1256, pp. 340-348, 2023.
• YAN, D. “Posthuman Creativity: Unveiling Cyborg Subjectivity Through ChatGPT”. Qualitative Inquiry 31, pp. 253-264, 2024. https://doi.org/10.1177/10778004241231923
  » https://doi.org/10.1177/10778004241231923
• ZHOU, E., LEE, D. “Generative artificial intelligence, human creativity, and art”. PNAS Nexus 3, pgae052, 2024. https://doi.org/10.1093/pnasnexus/pgae052
  » https://doi.org/10.1093/pnasnexus/pgae052
• ZUBOFF, S.The Age of Surveillance Capitalism“. New York: Public Affairs, 2019.
• ZWITTER, A., GSTREIN, O. J. (eds.). “Handbook on the Politics and Governance of Big Data and Artificial Intelligence”. London: Edward Elgar Publishing, 2013.

Edited by

• Responsible Editor:
  Mauro Luiz Engelmann.

Publication Dates

• Publication in this collection
  26 June 2026
• Date of issue
  2026

History

• Received
  28 May 2025
• Accepted
  20 Mar 2026