Rimini. A photograph circulates between chats and social media. It portrays a naked minor. Or, rather, the face is that of a real girl, while the body has been digitally processed. The image was created from real photographs by another minor, reported after the investigation Flying team conducted in mid-October. A handful of months earlier, in the Treviso area, the Postal police arrested a 57-year-old: 350 child pornography files on his computer, many of which were produced with theArtificial intelligence.
Again: another episode in the province of Teramo, where six teenagers created nude photos of classmates with AI, up to a law enforcement operation in Turin, when at the beginning of the year a Telegram channel through which synthetic child pornography content was circulating was stopped.
Emergency numbers between Italy and messaging platforms
«This is a real emergency» he comments Don Fortunato Di Notofounder ofMeter Associationwhich publishes the Child Victims report annually. «Through our monitoring we have highlighted 8,213 minor victims of child pornography images manipulated with artificial intelligence, identifying 115 active groups and bots among Telegram and Signal. In September alone we reported 592 victims. And among the complaints was that of a student who asked us for help in removing her contents manipulated by AI starting from material she had published on a well-known social network”.
The production of images, however, is only the first link in a chain that also includes their diffusion, exchange and, in some cases, sale. «Manipulative tools are very accessible today, and this is precisely the change in scale. Until a few years ago, to manipulate an image in a credible way, professional skills, time and software were needed. Today, a photo taken from a social profile, a few seconds and no skills are enough”, he explains Marco Camisani Calzolariexpert in artificial intelligence and digital risks. «There are nudity apps and sites, which present themselves as games or services for adults and are often paid for with credits, with a few euros» he continues. «Then there are bots on Telegram and other messaging platforms: you send a photo and the altered image comes back. And then there is the family that worries the most: i open source generative modelsfree, which you download and install on your computer. They run offline, without a company behind them, without a central panel to turn off and without it being possible to enforce the laws on what is generated. Added to these are the custom-trained models circulating in communities, built specifically to produce content that commercial services instead block.”
Global growth and the alarm of the Internet Watch Foundation
The consequence is that abuse becomes increasingly easier to carry out and difficult to stem. And that customized AI tools and models are often exchanged, including those based on already known victims. “Generative AI has significantly lowered the barriers to producing images of child sexual abuse, allowing perpetrators to create content at scale, quickly and easily. The cases we have identified are probably just the tip of the iceberg”, he explains to Panorama Kerry Smithmanaging director ofInternet watch foundation (Iwf), committed to combating the online dissemination of child pornography.
The numbers transmitted by the organization tell of a phenomenon that is growing at a dizzying rate: in the first six months of 2026 alone, 6,310 images of child sexual abuse generated through artificial intelligence were identified, 40% more than those recorded in the whole of 2025. In 98% of cases the protagonists were girls and boys, but the subjects even included newborns.
Around the world, the issue shows itself in its extreme gravity: in the United States, investigations and arrests linked to the production of synthetic child pornography material continue to emerge, and a recent report produced by Unicef And Interpol it emerges that at least 1.2 million children and adolescents declared that within a year they had had their images manipulated to produce deepfake sexually explicit.
Because a synthetic image causes real harm
But just because an image is artificial doesn’t mean the damage is any less real. “Both synthetic and genuine sexual images depicting children are harmful. The former can only be produced because images of actual child victims exist in the training data of AI systems. And all this leads to legitimizing the sexual abuse of minors”, explains one of the world’s leading experts on the topic, Clare McGlynnwho teaches law at Durham University. A few months ago, precisely to highlight the seriousness of the phenomenon, Unicef dedicated a position statement to the issue with the unequivocal title: Deepfake abuse is abuse. Which is to say: abuse through deepfakes is abuse, and there is nothing fake about the harm it causes.
To put a stop to the phenomenon it would be necessary to intervene simultaneously on technology, platforms and regulations. «What changes with AI is above all production capacity. You no longer need to have large image archives to generate massive amounts of new content. But this doesn’t just happen on the dark web, it also happens on ordinary platforms. It’s a structural problem, not something that can be solved by simply banning a program,” he reflects Walter Quattrociocchi full professor of Computer Science Wisdom of Romemember of the European Commission’s Artificial Intelligence Expert Group.
European rules, the role of schools and technological filters
«The most important problem concerns the application of existing obligations. A quick and effective application of Digital Services Act and of theAI Actwhich introduce important protections, would already make a significant difference. Platforms should then carry out effective assessments of the risks associated with their services. A first step would be the use of “hash matching“, i.e. the comparison of the fingerprints of already known illegal images to prevent their circulation”, reflects McGlynn.
Alongside the responsibility of the platforms, that of the school emerges, which should be invited to prepare emergency protocols and explain to younger people how sexually manipulating the image of a peer is not a joke, but an abuse. Furthermore, there is no shortage of prevention tools.
«Filters on textual requests, classifiers that analyze the image before and after generation, the recognition of the faces of minors associated with sexual content, the comparison with the “digital fingerprint” (hash) archives of already known child pornography material, managed by organizations such as Ncmec and Internet Watch Foundation are some of the tools we have available. But if the platforms don’t take it into account, the problem doesn’t stop”, adds Camisani Calzolari.
The legislative issue and the risk of the “liar’s dividend”
The legislative issue is more complex. At least in our country. «Even if our system does not start from scratch, the provisions were written before the advent of AI», he comments Giovanni Ziccardifull professor of Philosophy of Law at the University of Milan. «The Supreme Court is firm in protecting minors, but it is essential to understand the damage caused by an image, even if it is false. For a teenager who sees an artificially constructed sexual image circulating using his own face, the consequences on his reputation, personal relationships, dignity and psychological balance can be devastating. The fact that the episode depicted never happened does not make the damage imaginary.”
But there is also a further risk. «That the diffusion of synthetic images ends up weakening the credibility of authentic images. It is the so-called “liar’s dividend“, the liar’s advantage: those involved in events that actually happened can try to defend themselves by claiming that the Images were constructed by artificial intelligence. In this scenario, the technical competence of the investigators and the quality of the digital evidence become decisive”.
Kerry Smith of the IWF echoes this: «The rapid growth of synthetic child sexual abuse images demonstrates that voluntary initiatives are not enough. Companies developing AI systems should be obliged to integrate security into their products right from the design phase, while the government should introduce a regulatory framework capable of ensuring compliance with these obligations. The safety of children must become a fundamental requirement for the development of AI.” And not just on paper.



