AI-generated non-consensual intimate imagery has expanded sharply in recent years, with estimates rising from roughly 500,000 videos in 2023 to 8 million by 2025. That’s according to a new peer-reviewed study published in the Journal of Online Trust and Safety.

This research goes beyond simply measuring the growth of AI-generated sexual abuse. It also examines the internet infrastructure supporting websites that distribute this material.

Researchers identified 88 websites hosting AI-generated non-consensual intimate imagery, often referred to as AI-NCII. Of those, 38 were dedicated exclusively to the content. The study found that a small group of major technology and internet infrastructure providers repeatedly appeared across the websites’ technical infrastructure.

These findings have shifted part of the discussion around AI deepfake abuse, moving focus from the people creating the material toward the infrastructure that allows it to remain accessible online.

AI deepfake abuse reaches 8 million videos.

The study estimates that the volume of AI-generated non-consensual intimate imagery grew from around 500,000 videos in 2023 to 8 million in 2025. That represents a dramatic increase in just two years.

Researchers said the material overwhelmingly targeted women and girls. The websites examined primarily featured recognizable people from Hollywood and the Hindi film industry, along with K-pop artists and prominent feminists and activists.

This technology has also lowered the barrier to producing this type of abuse significantly. Researchers said a person with a single photograph available online can potentially become the target of fabricated intimate imagery.

That makes the problem genuinely different from earlier forms of image manipulation. The availability of generative AI allows abusive material to be created and distributed at a much greater scale than before.

Researchers identify 88 AI-NCII websites.

The research team examined 400 candidate web addresses during a six-week period in February and March 2026. After investigation, they identified 88 sites as actively hosting AI-generated non-consensual intimate imagery. Thirty-eight of those sites exclusively hosted this type of material.

The researchers then mapped the technical services these websites used, including hosting, content delivery networks, domain-name systems, domain registration, email, and advertising.

Five major technology providers repeatedly appeared across the infrastructure the researchers examined: Cloudflare, Google, Namecheap, WordPress, and Proton. The study describes Cloudflare as the most frequently recurring infrastructure provider in its sample.

Cloudflare appears across multiple infrastructure layers.

Cloudflare appeared particularly often because its services can cover several different parts of a website’s infrastructure at once. The study identified Cloudflare across hosting, content delivery, domain-name services, and analytics.

Researchers therefore argue that infrastructure providers represent an important potential point of intervention. If a website is dedicated to distributing AI-generated non-consensual intimate imagery, removing individual pieces of content may not be enough on its own. Cutting off key infrastructure could make it considerably harder for the operation to remain online.

The study does not, however, establish that the named infrastructure companies knowingly supported this abusive material. That distinction matters. The research maps which services were associated with the sites. It doesn’t, by itself, establish that every provider knew what individual websites were hosting or intentionally enabled the abuse.

Google and other technology providers also appear.

Google appeared in the study through services including security certificates and advertising. Namecheap was identified prominently in domain registration, while WordPress-related services and Proton’s email infrastructure also appeared among the websites examined.

These findings highlight how an abusive website doesn’t necessarily depend on just one company. Instead, it can assemble its operations from multiple layers of the internet ecosystem.

That creates a genuinely difficult enforcement problem. A hosting provider may not control a domain. A domain registrar may not control the content. An advertising provider may not host the website. A software provider may not operate the server. The result can be a fragmented technical structure where responsibility gets distributed across multiple companies at once.

93% of identified sites were searchable.

The researchers also examined how easily these websites could be discovered through Google Search. They found that 93% of the identified sites were easily discoverable through search, according to Lancaster University.

This finding adds another layer to the infrastructure problem. The material wasn’t necessarily confined to hidden networks or technically sophisticated underground systems. Some of the identified websites could be found through ordinary web searches.

That accessibility potentially increases both the reach and persistence of the abuse. It also means reducing distribution may require cooperation between multiple parts of the online ecosystem, including search, hosting, domain services, and content-distribution networks.

Researchers call for preventive technology.

The researchers argue that existing laws alone may not be sufficient to address the scale of AI-generated non-consensual intimate imagery. Their recommendations include stronger action by infrastructure providers against websites known to host the material.

They also point to perceptual hashing as one possible technical safeguard. This technique can create digital signatures of confirmed abusive material, so matching content can potentially be identified and blocked when it gets redistributed elsewhere.

The researchers also call on providers to block sites dedicated exclusively to AI-NCII and to suspend services for sites that continue hosting such material within broader offerings after receiving credible reports.

This approach would move some enforcement upstream. Instead of attempting to remove every copy after it spreads, platforms and infrastructure companies could try to prevent confirmed material from being redistributed in the first place.

A wider technology governance challenge

This study illustrates a broader challenge facing the technology industry. Generative AI can provide legitimate tools for image creation, editing, and automation. The same underlying capabilities can also be used to produce highly harmful material without a person’s consent.

The problem becomes particularly difficult when open technologies, third-party services, and mainstream internet infrastructure get combined. The study’s authors therefore focus not only on AI model developers but also on the wider technology stack that allows abusive websites to operate. That includes hosting, domain infrastructure, content delivery, advertising, email, and search visibility.

The scale of AI abuse is becoming an infrastructure problem.

The increase from an estimated 500,000 videos in 2023 to 8 million in 2025 illustrates just how quickly AI-generated non-consensual intimate imagery has expanded. But this new study adds another important finding: the distribution network can rely on ordinary internet infrastructure that most people use every day.

Researchers identified 88 active websites and found five major technology providers repeatedly associated with different parts of their technical operations. That creates a difficult question for the technology industry as a whole: how can companies preserve open internet infrastructure while preventing their services from being used to distribute clearly abusive material?

The researchers’ answer is to increase intervention at the infrastructure level, including better reporting processes, technical detection, and action against sites known to host AI-NCII. For the broader AI industry, this issue shows why responsible deployment can’t stop at the model itself. As generative AI becomes easier to use, the systems surrounding the technology, including hosting, distribution, search, advertising, and content moderation, are becoming equally important to preventing abuse.

AI Deepfake Porn Videos Grew From 500,000 To 8 Million In 2 Years, Study Says