A.I.-Generated Images of Child Sexual Abuse Are Flooding the Internet

A new and deeply alarming wave of artificial intelligence (AI)-generated child sexual abuse material (CSAM) is rapidly reaching unprecedented levels of realism, presenting significant challenges to global law enforcement and child protection agencies. These hyper-realistic images and videos, produced without the involvement of real children, are now so convincing that distinguishing them from actual abuse has become increasingly difficult—even for trained analysts.

Over the past two years, rapid advancements in generative AI have made it significantly easier for individuals with malicious intent to fabricate explicit and highly disturbing content featuring children. Organizations such as the Internet Watch Foundation (IWF) and the National Center for Missing & Exploited Children (NCMEC) are now sounding the alarm: 2025 is witnessing a sharp increase in this type of material, and it shows no signs of slowing down.

Newly released statistics from the IWF, a U.K.-based nonprofit dedicated to identifying and reporting online child abuse content, reveal that so far this year, 1,286 AI-generated CSAM videos have been flagged worldwide. This marks a dramatic increase from only two such instances reported during the first half of 2024. These figures underscore just how quickly the technology—and its misuse—is evolving.

According to analysts at the IWF, improvements in generative AI tools have resulted in smoother, more detailed videos. This progress has also been accelerated by collaborations among bad actors operating in the dark web—hidden areas of the internet where such material is created and exchanged beyond the reach of traditional oversight.

The emergence of highly realistic AI-generated videos compounds an already surging problem. In the United States alone, the NCMEC received an overwhelming 485,000 reports involving AI-produced CSAM in just the first six months of this year. This dwarfs the total of 67,000 reports for all of 2024. The explosive growth reflects not only the availability of the technology but also the increasing sophistication of those who exploit it.

Derek Ray-Hill, interim CEO of the IWF, described the situation starkly: “This is the canary in the coal mine.” He explained that AI-generated abuse material often includes both fictional images and photos of actual children, blurring the line between real and synthetic exploitation. “We’re witnessing a tsunami of this content,” he warned.

Although AI-generated CSAM still constitutes a relatively small proportion of the vast troves of child abuse material circulating online—millions of such images are reported each year—this new category is placing an additional burden on already overwhelmed law enforcement agencies. Each AI-generated image must be reviewed, assessed, and potentially investigated, diverting valuable resources away from cases involving real victims.

Legal authorities in the U.S. maintain that existing federal statutes prohibiting child sexual abuse and obscene content also extend to AI-generated material. These laws apply even when the content was created entirely through artificial means and does not involve actual children. Nonetheless, the novelty and complexity of AI have prompted state legislatures to take additional action.

Over the past few years, more than 36 U.S. states have enacted laws aimed specifically at criminalizing synthetic depictions of child sexual exploitation. Despite these efforts, legal experts acknowledge that the judiciary is only beginning to address the full implications of AI-generated CSAM, leaving many legal questions unresolved.

The root of this growing problem can be traced back to the explosion of generative AI technologies. Following the release of OpenAI’s ChatGPT in late 2022, other powerful tools soon entered the market—among them AI systems capable of producing lifelike images and videos. These tools have proven to be double-edged swords: while offering vast potential in fields like education, design, and healthcare, they have also raised profound safety and ethical concerns.

Unfortunately, much of the newly generated CSAM incorporates existing illegal content. In many cases, real abuse footage is repurposed into new formats, combined with publicly available images of children harvested from school websites, social media, and other platforms. The resulting content is then distributed through encrypted messaging apps, social networks, and underground forums.

In one notable case from December 2023, researchers at the Stanford Internet Observatory uncovered hundreds of CSAM examples embedded within a training dataset used by an early version of the image generator Stable Diffusion. Stability AI, which currently manages Stable Diffusion, stated that the dataset in question was compiled by a third party before it assumed full control over the model’s development. The company emphasized that it had not been involved in the training of that earlier version.

Recent technological breakthroughs have dramatically improved the realism of AI-generated images and videos. Earlier versions were often easy to identify due to unnatural features, such as distorted faces, an incorrect number of fingers, or choppy video transitions. Today’s tools, however, can produce visuals that are nearly indistinguishable from reality, avoiding many of the telltale signs that previously gave them away.

Last month, the IWF uncovered troubling discussions on a dark web forum, where users expressed admiration for a new batch of AI-generated CSAM videos. Forum participants praised the lifelike movements, highly detailed backgrounds—complete with furniture and wall decorations—and the presence of multiple figures engaging in disturbing and illegal acts.

John Shehan, a senior official at NCMEC, noted that approximately 35 technology companies are now submitting reports of AI-generated CSAM. However, he pointed out that not all companies are equally proactive. Those filing the highest number of reports tend to have more robust monitoring and safety protocols in place.

Among major tech players, Amazon—which provides AI services through its cloud platform—reported removing 380,000 instances of synthetic CSAM in the first half of 2025. OpenAI identified and took action on 75,000 cases. By contrast, Stability AI reported fewer than 30 incidents. The company insisted it has strengthened safety mechanisms and reaffirmed its commitment to preventing the misuse of its tools, particularly in relation to child exploitation content.

Both Amazon and OpenAI referred inquiries to their public transparency reports, where they outline their strategies for detecting and eliminating harmful content.

Unfortunately, criminals are now weaponizing these tools in more complex ways. According to a Justice Department official who spoke anonymously due to the sensitivity of the investigations, some offenders are using AI to fabricate explicit images of minors and then leverage them for blackmail. Others are creating “deepfake nudes”—digitally undressing real individuals using their photos—and sharing them via apps targeted at young users.

While materials involving actual children are unambiguously illegal, the legal landscape surrounding fully synthetic content remains murky. In March, a Wisconsin man accused of creating and distributing over 13,000 AI-generated images of child sexual abuse successfully challenged one of the charges on constitutional grounds. U.S. District Judge James Peterson ruled that the First Amendment may protect possession of obscene material created without real children—though he allowed other charges involving distribution and intent to share the material with minors to proceed.

Federal prosecutors said the man had attempted to send the synthetic images to a minor via Instagram, which subsequently alerted the authorities. “The Department of Justice considers all forms of AI-generated CSAM a serious and growing threat,” said Matt Galeotti, head of the DOJ’s criminal division.

As AI technology continues to evolve at breakneck speed, the tension between its capabilities and the societal harms it can enable is only growing. Child protection experts, policymakers, and technologists now face the urgent challenge of working together to ensure that the benefits of AI are not overshadowed by its potential for abuse.