In May 2025, the TAKE IT DOWN Act took effect in the United States - the first federal law against deepfakes and other AI-generated intimate images created without a person's consent. It criminalizes the distribution of "porn revenge" (intimate photos or videos published without the consent of the person depicted), including materials created by artificial intelligence, and requires social networks and websites to promptly remove such content upon request.
However, a new study from Brown University shows that this law and the state laws adopted before and after it do not cover the entire digital ecosystem that makes this type of AI-enabled sexual violence possible. The work was presented on November 7, 2026, at the ACM Conference on Equity and Access in Algorithms, Mechanisms and Optimization.
The study was led by Michelle Ding, a computer science graduate student at Brown University's Center for Technology Responsibility, Reimagining and Redesign. "The key problem is that deepfake abuse is not a single image, model, website or app, but an entire ecosystem," Ding says. "To protect vulnerable communities from harm and emotional suffering, we first need to see that ecosystem clearly."
The researchers compiled a taxonomy - a classification system - of 11 technology categories covering five roles: image creation (the AI models themselves, interfaces, training data), distribution (social networks and messaging apps), detection and replication (search engines, app stores, communities), infrastructure and support (developer platforms such as GitHub and Civitai), and monetization (payment systems).
As an example, they analyzed the ecosystem around Grok on the social network X. Users could get an intimate image simply by tagging Grok in a post with a request to "put a bikini on her" or "take off her clothes." The bot posted the image directly in the comments. Then a community emerged on Telegram where people taught others how to bypass the restrictions of Grok's paid version and create even more explicit material.
After analyzing 64 state laws and federal acts, the researchers found that more than 92% of them target individual creators and distributors, with almost no effect on the technology infrastructure. According to the authors, this creates an endless game of "whack-a-mole" - a chase after individual images across the internet. "Grok generated more than 6,000 images per hour in December of last year. Attention needs to shift to the technical infrastructure that allows abuse to happen and generates profit," Ding emphasizes.
Co-author Harini Suresh, an assistant professor of computer science, adds: "Right now all measures are very reactive - images are removed once they have already spread. There is a gap on the creation side. Can we motivate AI researchers to think about how to prevent models from creating such images in the first place? This is a difficult task, but perhaps we can help put it on the agenda."
The researchers hope the ecosystem map will help legislators and companies find pressure points. A companion website, created with the participation of students Haia Bugshan and Maya Mishevska, presents the technologies in an interactive format while avoiding sensationalism. "When we talk about the ecosystem, it sounds abstract. But by showing familiar tools and websites, we make it tangible," Bugshan says.