From Style to Sovereignty

Abstract

Since generative artificial intelligence entered large-scale cultural production in 2022, the protests, boycotts, lawsuits, union actions, technical defenses, and platform-governance measures surrounding it have often been grouped under the broad label of “anti-AI.” Yet these events do not constitute a unified camp of technological conservatism, nor can they be reduced to a simple narrative of humans resisting machines. Rather, the anti-AI movements of 2022–2026 are better understood as a history of shifting control over the conditions of generation. Visual artists first experienced a loss of control over the extraction and invocation of style; writers confronted the dilution of rights once texts entered training systems; performers and musicians faced the possibility that voices, faces, gestures, and performances could be detached from their bodies and made to work in their absence. Copyright capital, platforms, and state actors subsequently entered the same conflict, seeking to regulate model capabilities, intellectual-property boundaries, data flows, and infrastructure. In this process, “anti-AI” moved from the affective resistance of individual creators toward an institutional contest among differently positioned actors over the conditions under which generation is allowed to occur.

This essay interprets that transformation as a historical movement from “style” toward “sovereignty.” Sovereignty here does not refer only to the state. It names the authority to determine the callable status of one’s works, body, voice, identity, data, intellectual property, and technological infrastructure. The 2022 ArtStation “No AI” occupation, legal claims over copyright, attribution, and opt-out mechanisms from 2023 onward, technical defenses such as Glaze and Nightshade, WGA and SAG-AFTRA negotiations over digital replicas and performance rights, the institutional incorporation of AI art by the art market in 2025, and the 2025–2026 conflicts involving Disney, the MPA, Seedance, and other rights holders together reveal a structural shift: refusal no longer remains merely an attitude; it increasingly enters contracts, platforms, technical systems, law, and infrastructure. The historical significance of anti-AI movements therefore lies less in whether they succeed in stopping artificial intelligence than in how they compel cultural production to ask a more fundamental question: who has the authority to decide how cultural information may be used, and through what institutions is that authority made durable?

Keywords: generative AI; anti-AI movements; creative labour; style; digital replicas; control; infrastructure; sovereignty

1. From Technological Fear to a Crisis of Control

Anti-AI movements do not represent a single technological fear, nor are they naturally equivalent to a new Luddism. Since 2022, the actions gathered under the label “anti-AI” have emerged from radically different social positions. Illustrators have opposed the unauthorized extraction and imitation of style; writers have challenged the untraceability of texts once absorbed into training systems; actors and voice performers have resisted the digital replication of voices, faces, and performances; musicians have questioned voice cloning and its revenue structures; copyright holders have attempted to re-establish the boundaries of intellectual-property worlds; while platforms and states have increasingly turned toward questions of data access, model capability, and infrastructural control. These actors share a vocabulary about artificial intelligence without sharing the same interests, risks, or power. “Anti-AI” is therefore better understood as a series of defensive responses produced when differently situated actors perceive that generative systems are redistributing control, rather than as a politically unified community.

Control, in this context, is not an abstract metaphor for power. It concerns concrete authority over the conditions of generation: who decides whether a work may enter a training corpus; who may invoke a style; who may reproduce a face, a voice, or a performance; who may authorize a model to generate characters and narratives from an existing IP universe; and who may set the boundaries of these capacities through platform rules, contracts, defaults, technical barriers, and legal procedures. The distinctive force of generative AI lies not simply in its ability to produce images, texts, and sounds, but in its conversion of previously dispersed cultural traces—works, styles, bodies, voices, behaviours, and identities—into resources that can be learned, invoked, recombined, and reproduced. In that conversion, elements once bound to a creator’s personhood, labour, and history are recoded as operational capacities. “Anti-AI,” at its most basic level, begins as a refusal of that operability.

The history of anti-AI is therefore not a linear story of machines “invading art,” but a history of translation around generative control. At first, creators described their experience through the affective language of theft, imitation, humiliation, and replacement. They were then compelled to translate those injuries into the institutional languages of copyright, attribution, data authorization, opt-out mechanisms, collective agreements, technical interference, and platform governance. By 2025–2026, the conflict had moved further into capital and infrastructure, where the self-defense of precarious creators and the governance demands of powerful rights holders could both appear under the same semantic umbrella of “anti-AI.” It is here that cybernetics and the desire for control begin to overlap: generative systems depend upon technical control over information, conditions, and outputs, while the politics surrounding them increasingly concern the reappropriation of that capacity by competing social actors.

2. Style: Aesthetic Injury and the Crisis of Subjecthood in 2022

If this history requires a point of departure, it did not begin with national policy, labour agreements, or large-scale copyright litigation, but with the apparently private and difficult-to-define object of style. Visual artists were among the first groups to experience the crisis of control produced by generative AI. They discovered that habits of composition, brushwork, palette, figure treatment, and visual language developed over years could be rapidly invoked through generative systems; artists’ names themselves could enter prompts and function as retrieval cues for a visual effect. Once the creator’s name shifts from authorial identity to generative condition, what occurs is not merely the copying of works in a traditional sense, but a more complex transformation of subjecthood: the artist’s history, labour, and visual persona are compressed into a function callable by others.

This helps explain the emotional intensity of the 2022 ArtStation “No AI” image occupation. It was not primarily a rigorous copyright argument, but a visual act of obstruction. Artists flooded an image platform with anti-AI imagery, inserting the sign of refusal into the exhibition system itself and interrupting the frictionless circulation of AI imagery. Its significance does not depend on whether those protest images count as art. What matters is that the platform itself became the site of struggle and refusal became visible as an image. Compared with the later institutional languages of contracts, litigation, and technical tools, the 2022 protests remained largely symbolic and affective: their keywords were theft, imitation, replacement, humiliation, and betrayal by the platform. What was threatened was not simply a particular work but the creator’s understanding of the relation between style and personhood.

For precisely this reason, the disputes of 2022 could not be fully absorbed by conventional copyright language. Copyright is well equipped to address works, copies, permissions, damages, and ownership, but is far less capable of describing experiences such as the dilution of style, the conversion of a name into a prompt, the compression of long-term training into model capability, or the statisticalization of creative persona. Visual artists confronted an aesthetic injury for which legal language remained underdeveloped. What appeared to be taken was not always an identifiable object, but a sense of control over the continuity of one’s own practice. Style thus ceased to be merely a formal property or a market signature and became one of the first sites at which the politics of generative control became visible.

3. Claiming Rights: From Aesthetic Injury to Institutional Language

After 2023, the structure of anti-AI movements changed markedly. As lawsuits, copyright disputes, data-authorization mechanisms, and opt-out tools proliferated, creators began converting injuries previously expressed in affective terms into institutional claims. The shift from “do not steal my style” to “recognize my labour, my works, and my rights over data” is not merely a stronger version of the same statement. It signals a transformation in the logic of resistance. The former centres the injured subject; the latter attempts to compel law and platforms to recognize that injury as something that can be described, delimited, and compensated institutionally.

This was one of the central contradictions of 2023: the modes of harm experienced by creators were more complex than the modes available to existing law for describing them. Style is not naturally equivalent to a copyrighted work; statistical learning is not identical to conventional copying; and training often occupies an intermediate zone between data acquisition, model optimization, and final generation. Anti-AI movements therefore had to remake their own language, translating claims about creative dignity, aesthetic personhood, and the history of labour into the vocabulary of copyright, data authorization, attribution, withdrawal rights, and permission to train.

This translation expanded the institutional force of anti-AI claims while simultaneously revealing their limits. Law may determine whether a work has been copied, yet remain unable to decide whether a style belongs to a subject; it may define an authorization relationship, yet struggle to evaluate the transformation that occurs when a person’s name becomes a prompt; it may calculate damages, yet find it difficult to describe how years of creative experience are reorganized inside a model as an indefinitely callable capability. The anti-AI movements of 2023 therefore did not resolve the problem of style. They forced it into a larger institutional field. Style shifted from a private aesthetic attribute into a contested object that rights language could not entirely contain, preparing the ground for later conflicts over bodies, voices, labour, and digital replicas.

4. The Body: From Rights in Works to the Separability of Personhood

As the conflict expanded into the labour of actors, voice performers, musicians, and game performers in 2023–2024, the object of anti-AI resistance began to move from the work toward the body. For visual artists, generative AI had first made it possible for style to be learned and invoked apart from particular works. For performers and musicians, the problem was more immediate: voices, faces, gestures, performances, and elements of persona could be detached from the originating body and made to continue working in its absence. The artist fears that “my style no longer needs me”; the performer confronts the possibility that “my body no longer needs me.”

This shift moved the politics of anti-AI beyond authorship toward personhood and labour. Voice cloning, digital replicas, motion capture, and synthetic performance are not adequately described as machines “taking jobs.” More fundamentally, they produce the separability of the body. Voice, face, posture, and performance—capacities previously bound to the living time of a particular worker—are transformed into assets that can be stored, licensed, copied, and repeatedly invoked.

The significance of actions by the WGA, SAG-AFTRA, and related labour organizations therefore lies not only in compensation but in the re-establishment of control over the relation between the body and its digital replica. Once an actor’s performance can be datafied and regenerated in the actor’s absence, labour is no longer confined to the concrete site of recording or production; it can become the repeated invocation of previously captured bodily information. Control therefore enters the domain of personhood: whether replication is permitted, for how long, in what context, whether renewed consent is required, and how revenues are distributed all become more precise and consequential questions than whether “AI should be used.” Anti-AI thus expands from a dispute over works and styles into the governance of callable personhood.

5. Institutionalization: When Artificial Intelligence Enters the Art System

By 2025, the object of anti-AI resistance shifted again. Earlier disputes had centred on the extraction of works, styles, and labour by platforms and model companies. As AI art entered auctions, exhibitions, and institutional frameworks, conflict increasingly targeted the mechanisms by which the art world confers legitimacy. The significance of Christie’s AI auction lies not simply in the fact that “AI art was sold,” but in the conversion of generative mechanisms previously contested by creators into objects that could be collected, priced, and traded within the conventional art market. For opponents, this suggested that injury was no longer produced only by technological systems; it could receive renewed validation through art institutions themselves.

The art system has long possessed a powerful capacity for absorption. Conflicts, critiques, and dissident practices that begin at the margins of institutions can, once admitted into exhibition, collection, and market systems, be recoded as objects to be viewed, traded, and historicized. The institutionalization of AI art follows a similar pattern. As auction houses, museums, and other cultural organizations begin treating generative AI as a medium that can be incorporated into established systems of value, the question shifts from “does this technology belong to art?” to “who has the authority to legitimize it?” In this sense, the anti-AI sentiment of 2025 exceeded rejection of a tool and turned toward the possibility that art institutions themselves were repackaging unresolved disputes over training, consent, and labour as artistic innovation.

Christie’s should therefore be understood not as an isolated market event but as an institutional turning point. Generative AI no longer appeared only as an external technological force acting upon the art world; it began to participate internally in the distribution of value. The object of anti-AI resistance expanded accordingly. Creators no longer asked only that model companies change their training practices, but also questioned how exhibitions, auctions, and cultural institutions participated in the normalization of AI art. It is here that anxiety over legitimation emerges.

6. Capital: From the Protest of the Weak to the Governance of the Powerful

The disputes involving Disney, Universal, the MPA, Seedance, and related actors in 2025–2026 exposed one of the most important differences hidden inside the term “anti-AI”: the difference in power among those who resist. Visual creators in the ArtStation movement largely occupied a weaker position relative to platforms and model companies; they sought a minimum degree of control over their own styles, works, and labour. Major film studios and copyright organizations faced a different kind of risk: generative models threatened to cross the boundaries of established IP universes and weaken long-standing systems of proprietary control. Once powerful copyright holders began restricting, litigating, or negotiating model capabilities, “anti-AI” could no longer be described simply as creators resisting technological capital from below.

Two distinct logics of control must therefore be separated. One is the defensive protection of subjects attempting to prevent unauthorized use of their works, bodies, and identities. The other is the effort of established capital to preserve monopolistic control over cultural assets, character worlds, and commercial licensing. The former is a politics of protection; the latter approaches the re-governance of generative infrastructure. Both may speak the language of copyright, permission, and licensing, and both may demand constraints from model companies, yet their positions in the structure of power are fundamentally different. To collapse them under the single heading of “anti-AI” would flatten the history. A more precise account must recognize that anti-AI movements include both the self-defense of precarious creators and the re-regulation of generative infrastructure by powerful copyright capital.

This development makes “sovereignty” a useful concept for interpreting the conflicts of 2026. Sovereignty here is not merely an extension of state sovereignty. It concerns the final authority to determine the permissible scope of generation. Illustrators seek control over style; actors over body and voice; writers over the conditions under which texts enter training systems; musicians over voice and revenue; film companies over the boundaries of IP universes; model companies over training data and generative capability; and states increasingly over compute, data, models, and infrastructure themselves. At this level, the politics of generative AI no longer derives primarily from philosophical questions about whether AI is conscious, but from a more practical question: who gets to define what may be generated?

7. Refusal: From Political Gesture to Cultural Form

Running parallel to law, capital, and institutional governance is another, more dispersed but equally important historical line: the formalization of refusal. From ArtStation’s “No AI” images and the silence of protest albums to technical defenses such as Glaze and Nightshade and user boycotts such as QuitGPT, anti-AI action increasingly operates not only by speaking but by refusing to upload, authorize, subscribe, generate, watch, collaborate, or allow models to train smoothly. Non-participation becomes an action structure in its own right.

To understand refusal as a cultural form does not require treating every act of resistance as art. QuitGPT is not an art event, and Glaze and Nightshade are first of all technical tools. Their significance lies in what they reveal together: when access to technology no longer guarantees autonomy, withdrawal, interference, withholding permission, and restricting circulation acquire cultural-political significance. Generative systems expand through accessible data, callable content, continuous user participation, and frictionless platform operation. Under those conditions, non-participation becomes a negative action directed at infrastructure. It does not create a new image, but changes whether images can become training material; it does not create a new platform, but changes what a platform may treat as available by default; it does not create a new model, but attempts to constrain what models may absorb and generate.

If the “No AI” images of 2022 were still primarily a symbolic visual refusal, by 2026 refusal had increasingly entered concrete institutional structures: opt-out mechanisms, licensing systems, collective agreements, anti-scraping tools, model restrictions, platform defaults, and commercial licensing. Anti-AI movements thus reveal a crucial historical transformation. Refusal no longer stops at negation; it begins to redesign the conditions of generation. Its aim is not necessarily to abolish artificial intelligence, but to force it to operate under newly contested conditions.

8. From Style to Sovereignty: The Formation of a Politics of Generative Control

Looking back across anti-AI movements from 2022 to 2026, one can discern a historical arc that was not designed in advance but has gradually become visible. Conflict begins with style, passes through works, data, labour, bodies, and identities, and eventually reaches intellectual property, platforms, models, and infrastructure. These phases do not replace one another; they accumulate. The problem of style does not disappear when labour agreements emerge, nor do bodily rights become irrelevant once powerful copyright actors enter the conflict. Rather, generative AI continually expands the domain of cultural information that can be operationalized, transforming increasing amounts of what had previously belonged to subjectivity, professional convention, or institutional tacit knowledge into objects that can be trained, copied, generated, and traded.

“From style to sovereignty” therefore does not describe a neat evolutionary sequence. It names an expansion in the scale of the object under dispute. At first, creators ask models not to invoke a style; then they ask law to recognize boundaries around works and training data; unions seek control over digital replicas and performance rights; copyright capital demands that models conform to licensing systems; and platforms and states compete over the governance of data, model capability, and infrastructure. Style matters not because it is the final issue, but because it was one of the first places where this broader crisis of control became perceptible.

In this sense, anti-AI movements reveal not whether “humanity accepts artificial intelligence,” but the political structure of generative culture. AI has never been merely a neutral medium. It depends on the acquisition of data, the training of models, platform distribution, copyright boundaries, the organization of labour, and the control of infrastructure. As these conditions become visible, anti-AI shifts from an affective reaction to technology toward a struggle over the conditions of generation themselves. Technical control and the social desire for control mirror one another: generative models expand their capacity to encode and invoke cultural information, while different actors attempt to lock that capacity back within their own regimes of permission, property, identity, and institutional authority.

The most significant aspect of anti-AI movements from 2022 to 2026 is therefore not whether they ultimately “stopped” artificial intelligence, but how they made the problem of control in cultural production visible. They force artists, workers, platforms, capital, and states to answer anew: once a work, a style, a body, a voice, an IP universe, or an entire cultural infrastructure enters a generative system, to whom does it still belong, and what does belonging mean under those conditions? The history of anti-AI is ultimately not a history of humans against machines, but a history of how cultural information is invoked, refused, authorized, copied, governed, and reappropriated.