On May 14, 2026, MIT Technology Review published a long-form investigation documenting how adult performers' bodies, scenes, and identities are being scraped from existing pornography to train AI image and video generators — without consent, compensation, or any practical way for performers to confirm whether their work was used. The report estimates that "more than 10,000 terabytes" of online pornography likely contributes to the training of nudify apps, AI-companion image generators, and synthetic-video tools, and traces concrete economic harm flowing to performers whose deepfaked likenesses are now competing with their own paid content.

Why It Matters

This is the first piece of major mainstream coverage to articulate the AI-training-data problem from the angle of the adult performers whose careers are most directly affected. Most public discourse around nonconsensual deepfakes has centered on victims who were not adult performers (Taylor Swift, the Tennessee teens suing xAI, Almendralejo schoolgirls); the MIT Tech Review framing reframes the same pipeline as an industry-displacement story for the people who literally created the underlying content. For sex tech and content-platform investors, the implication is that the same generative-AI tailwind hyped as bullish for the broader content economy is structurally bearish for the working performers and creator-economy platforms that depend on authentic content as a moat. Expect adult-industry trade associations (FSC, ASACP) and creator-economy platforms (OnlyFans, Fansly, ManyVids) to begin pushing legislative and litigation strategies aimed specifically at AI training data — distinct from the existing nonconsensual-deepfake-of-identifiable-victim track that the TAKE IT DOWN Act addresses.

The investigation's central source is "Jennifer" (pseudonym), an adult performer and psychotherapist who discovered AI-generated content using her body and face on multiple platforms. The reporting also documents AI-duplicate hosting platforms like Spicey AI (now defunct), nudify apps including Crushmate, and Grok's "Imagine" video mode, which independent analyst Genevieve Oh has called "unmistakably the largest nonconsensual synthetic nudity generator" in the world. UC Berkeley digital forensics professor Hany Farid and MIT computer science researcher Stephen Casper both characterized the training pipeline as operating as a "black box" — performers cannot prove their work was used in a specific model, which makes both consent enforcement and copyright claims structurally difficult to pursue.

The economic mechanism is the part that most directly affects the adult industry's business model. Reba Rocket, chief operating officer of Takedown Piracy, told MIT Tech Review that AI-generated alternatives are beginning to "undercut demand for authentic content" because they cost a fraction of producing scenes that require filming equipment, locations, paid performers, editing, and marketing. The report quotes attorney Corey Silverstein, who specializes in adult industry law, on the practical inadequacy of existing tools: copyright applies to specific works, not to the underlying physical likeness; right-of-publicity claims vary by state; and the federal TAKE IT DOWN Act — whose platform compliance deadline is May 19, two days from this story's publication date — applies to nonconsensual intimate imagery of identifiable individuals but not to AI-generated synthetic bodies trained on aggregate scraped data without a one-to-one identifiable subject.

Anne Craanen, a gender-based violence researcher at the UK's Institute for Strategic Dialogue, framed the broader stakes in the article: AI training pipelines that quietly absorb adult performers' work create a structural feedback loop in which the most exploited segment of the content economy underwrites the technology that will eventually displace it. The report names Meta/Instagram, Apple, Google Play, X (Grok), and a handful of specialized hosting services as the primary surfaces where the resulting synthetic content circulates — though most have content-moderation policies that nominally prohibit non-consensual deepfakes, enforcement remains inconsistent.

Sources

Update — 2026-05-17

Initial entry — story first created.