With the advent of generative AI, it is important to differentiate between “images of violence” and “images of generative violence”. This is not an ontological distinction but rather a social one, with anthropological and political ramifications. Society has used “images of violence” since long before the invention of photography. As a form of iconography, representations of violence have served political, religious, and personal purposes.² By contrast, “images of generative violence” produced by AI are not mere representations—they constitute a new form of violence that can cause psychological trauma and contribute to other forms of physical violence, much like verbal violence or violent games. However, the violence of generative AI images is a more insidious phenomenon. The ability to create sophisticated images, rapidly tailored to each viewer’s context, makes them highly powerful and manipulative. These images generate violence and, ultimately, amplify and normalized the desire for it. Although Generative AI-Violence is purely visual, its pervasiveness and deceptiveness can have serious repercussions for individuals and for society at large.

Generative AI-Violence is a mass violation, open to all and directed against each of us. It differs from past exposure to cruelty and atrocity through imagery. Now, everyone is potentially either a casualty or an aggressor capable of inflicting violence from a distance. The violation of individuals is now produced for the masses and disseminated rapidly and widely, which makes it even more dangerous. In addition, Generative AI-Violence is algorithmically engineered to exploit our most personal vulnerabilities through the large-scale accumulation and computation of personal data and information relating to social controversies. Generative images tend to become increasingly violent by themselves through learning from violent prompts of a violent society, as well as through learning from psychological vulnerabilities and social abuses embedded in cultural biases and conflicts.

A feedback loop is emerging between “Machine Vision” and the “Envision Machine”.³ In a cybernetic sense, this feedback within AI allows the generative images of the “Envision Machine” to improve continuously through its capacity to classify images via “Machine Vision”. Generative AI is learning the capabilities of its own images and the effects they have on humans. The generative “Envision Machine” progressively influences our personal imagination, thereby transforming our culture.

These days, AI is being used to repress society not through force, but by seducing us with cute, monstrous, sexual, and extreme images. Generative AI can produce and optimize visuals designed to attract viewers through violence and abuse. Consequently, a violent collective sensibility and imagination are being reproduced and desensitized throughout society. AI is training citizens to desire violence, and, through a feedback loop, citizens are also training AI to become more abusive by embedding within it an increasingly biased, hateful, and violent society.

This new media phenomenon must be distinguished from the imaginary and imagery produced by other visual media such as photography, cinema, or video games; although these media forms have played an important role in establishing a visual culture of violence within society. The advent of the tabloids in the 1950s, bloody films in the 1970s, and video games in the 1990s all represent stages in a cultural evolution that normalized increasingly aberrant and violent imagery. However, Generative AI-Violence is different. It entails a more direct form of violence that can target an individual’s dignity and then broadcast that violation in real time. Such personal violence can even be automated for each individual and recursively self-reinforced. AI-generated images can be manipulative not only in an iconographic sense, but also in a deeply physiological one. They intimately target the subconscious by leveraging personal data, producing profoundly abusive forms of violation.

It is also important to consider the interfaces and devices through which AI-generated violent images are delivered. These images can assault individuals in both their intimate and social lives. Victims of Generative AI-Violence are often violated within their own private spaces, in isolation, while platform interfaces offer little opportunity to seek help or support. In such an environment, violent images are designed to be highly addictive and to continually capture our attention, which is then monetized by platforms and, in some cases, by users themselves. Content creators and social media platforms have an economic incentive to publish and produce provocative and shocking content.⁴ The greater the controversy and violence, the more revenue they can generate through internet traffic, advertising, subscriptions, and the mining of viewers’ reactions. At the same time, platforms practice censorship by subtly filtering and arranging feeds to distort reality, rather than simply blocking content, which is an even more valuable asset when it is controversial. The environment in which AI-generated violent images circulate is crucial to understanding their politics, functions, and consequences.

Nevertheless, it is the cultural misinterpretation of “images of generative violence” that ultimately allows them to exist and circulate. This is not merely a technological or media phenomenon, but also a cultural one that legitimizes and justifies such violence and abuse. To address this misinterpretation, it is important to distinguish between morality and ethics when judging the role of these images.⁵ Depictions of sexual abuse and acts of violence have been assimilated into Western culture through literature, cinema, comics, and photography.⁶ Such representations rightly stand as part of a cultural revolution against the moralism of religions and authorities that, for centuries, imposed cultural norms through censorship and social rules. However, revolutions against morality should be distinguished from violence employed unethically. Free expression concerning mischief, perversion, sadism, horror, and brutality, even to the verge of the unthinkable, must still be understood as a liberation from taboo images and sexual repression. Such freedom of expression is one of the fundamental pillars of modern democracy and must remain so. Ethical violent images should therefore be permitted, while being distinguished from unethical “images of generative violence”.

For instance, the mass violation of women and children has nothing to do with free speech. Today, liberalism is often confused with libertarianism, not only in politics and technology, but also in culture and art.⁷ Ethics is not morality, just as fundamental rights are not simply protections of individual freedoms. For example, freedom is confused with allowing minors to be exposed to toxic content and platforms;⁸ the right to show nipples on social media⁹ is confused with the right to sexualize someone’s image without consent; and pornographic content is confused with the right to humiliate someone under the guise of irony. Yet such acts are neither fun, nor sexy, nor just mean. They are blatant violence. This form of violence is directed not only at women, children, minorities, or other vulnerable individuals; it is a form of mass violence that potentially affects everyone. It may even be understood as a form of social terror, insofar as it can occur at any moment, without warning.

It is also important to differentiate between the subjects of this violence. Generative AI-Violence can be used to abuse, stigmatize, and humiliate real individuals. These are not images of fictional personas, actors playing roles, or abstract public figures. Generative AI-Violence is not fantasy, fiction, or a game. Nor is it merely voyeurism or depravity. Rather, it is violence directed at real people with real lives who have neither given consent nor desired the generation of such images of themselves.¹⁰ Even when the subjects are well-known celebrities or politicians, we should still consider the personal and political violence that AI-generated images can inflict.¹¹ These images can be so harmful that, even when they appear to be satire, they may still conceal violent intentions. We may therefore need to establish new standards for parody. Even seemingly simple AI-generated violent images can produce harm within the broader context of the highly powerful and manipulative media environments in which they circulate.¹² For instance, so-called fake news has already contributed to social unrest, mass killing, and lynchings. Inevitably, generative AI can infect images with new forms of violence, producing both direct and indirect consequences for individuals and for society.

The misunderstanding and misconception of the role of such images and devices reflect a broader culture of tolerance towards violence, which is mirrored in politics and ultimately in the deregulation of such content and technologies. Generative AI-Violence should be understood as a political instrument. These abusive AI-generated images can serve to destabilize democracies by degrading democratic values. The establishment of such cultural domination forms part of a broader far-right political agenda, which has increasingly been supported by major technology firms that generate and promote such content in the first place. This collusion of political and commercial interests is reflected in recent lobbying efforts aimed at deregulating content moderation, data privacy, and AI safety laws. These agendas are advanced by big-tech firms and far-right political groups in concert.¹³ Yet the political function of Generative AI-Violence is not limited to the establishment of cultural hegemony; it also serves to produce fear and intimidation among citizens. It is a weapon that can be directed against anyone through the narrative that AI is uncontrollable and impossible to halt. In this sense, our images are already being weaponized, as society increasingly tolerates and justifies the existence of such violence. We must avoid a future shaped by a "Violent Machine" that learns violence and reproduces it—and that could eventually, as generative forms of "artificial general intelligence"¹⁴ develop the capacity to enact physical forms of violence. When Generative AI-Violence moves beyond images and is deployed in military, policing, and politics through robotics, weaponry, and critical infrastructures, it risks annihilating society through brute physical force.¹⁵

Today, a greedy, destructive feedback loop between politics and technology is rapidly evolving, eroding civic values and democracies. Art can expose evidence of AI abuse and violence to a broader public while inspiring constructive responses.¹⁶ It is important to develop cultural instruments capable of interpreting the complexities of contemporary society. Generative AI will inundate society with images of violence, and artists and cultural producers will increasingly need to guide public understanding of the social role of such images and of the fine distinctions between them. It remains essential to differentiate between "images of violence" and "images of generative violence", starting from the rejection of violence in all its forms. Ultimately, this is not just a question of morality or ethics. Above all, it is a matter of human rights.