How 4chan's Deepfake Communities Turn Harassment into a Collaborative Game

New research from the Institute for Strategic Dialogue reveals that the creation of nonconsensual intimate imagery — fake explicit images of real people — operates as a team activity that builds social bonds within online misogynist communities. The study found that 4chan's /r/ board serves as a central hub for coordinating these activities.

The researchers analyzed thousands of posts from December 2025 through March 2026 on the platform's 'adult requests' board. What they found was a structured system: users post requests for fake explicit images of specific women, then community members — who call themselves "wizards" — create and share the synthetic material.

The Social Machinery Behind the Harassment

The research shows that people who create this nonconsensual imagery receive praise and recognition from their community, which reinforces the behavior. This is the same kind of social feedback loop that drives other forms of group harassment. The system turns what could be isolated acts of technological abuse into a collaborative exercise where people take on different roles — some request content, others create it, and others distribute it.

The vast majority of people targeted are women, a pattern consistent with what researchers have documented across other deepfake pornography platforms. Because requests come from the broader community, multiple people develop an emotional investment in the harassment campaign.

When the research team examined how this content spreads, they found that material created on 4chan's request board moves onward to messaging apps like Telegram and Discord. This distribution extends the harm far beyond the original creation point.

The Technology Behind the Imagery

The synthetic images rely on a technology called Generative Adversarial Networks, or GANs — essentially two competing AI systems that become good at swapping one person's face onto another image or video. These tools have become progressively easier to use. You no longer need machine learning expertise to create convincing fake explicit images; increasingly user-friendly software allows people with basic technical skills to participate.

Academic research has quantified the problem: 96% of deepfake videos online are nonconsensual pornography, with women disproportionately represented among victims. The request-response model documented on 4chan shifts the pattern from individual creators working alone to a distributed network where technical skills function as a shared community resource aimed at targets chosen by the group.

How Law and Platform Enforcement Have Responded

Both legal and platform-level interventions have attempted to address nonconsensual synthetic imagery, with uneven results. The TAKE IT DOWN Act and the removal of major deepfake sites like MrDeepFakes did reduce production volumes, but the research suggests that deepfake pornography is proving difficult to suppress completely.

Several countries have moved to criminalize the creation of such images. Britain recently introduced laws specifically targeting creation, not just distribution. Sweden has similar frameworks in place. In the United States, the Department of Justice has prosecuted individuals under the Take It Down Act, with cases covering both adult-targeted and minor-targeted synthetic content.

A recent example: the city of Baltimore sued Elon Musk's company xAI over nonconsensual images generated by the Grok chatbot, opening a new legal front by targeting AI companies whose general-purpose models can be misused this way.

Federal agencies including the FBI have issued public warnings about malicious actors using deepfakes for harassment and sextortion schemes. These materials circulate widely on social media, forums, and pornographic websites.

What This Pattern Reveals About Online Harassment

The broader context here is that we have seen this cycle before. Coordinated trolling campaigns, doxxing operations, and swatting networks all followed a similar trajectory: they built community through shared participation in targeting others, technical barriers gradually lowered to allow more people to participate, and each generation of harassment technology became more effective and widespread. The request-response deepfake model is the latest iteration of this pattern, now applied to synthetic media.

The Challenge for Platforms and Policymakers

Traditional content moderation — removing prohibited material after it appears — is not enough to address what this research documents. Because creation is collaborative and spread across multiple platforms, meaningful intervention has to disrupt the entire system: the request infrastructure, the social reward structures that recognize creators, and the networks that distribute the content.

Platform operators must now tackle established social communities, not just filter individual pieces of content. The fact that creation and distribution happen across multiple platforms — 4chan, Telegram, Discord — makes unified enforcement even harder.

The Institute for Strategic Dialogue's research supplies empirical evidence for policy discussions around AI safety, content moderation, and how to protect people from this form of digital harm.