AI-Generated Adult Content: The Business and Policy Reality

Generative AI's most contested application isn't a research paper or a product demo — it's the fast-growing category of tools that generate sexual imagery of real people without consent, often marketed as "nudify" or "undress" apps. In 2026, this stopped being a purely ethical debate and became a live legal and business-risk problem, with new federal and state laws, coordinated app-store enforcement, and a foreign regulatory regime (the UK's) that now criminalizes creation itself, not just distribution. This piece looks at what actually changed, who is liable, and what it means for anyone building or deploying generative AI image and video tools — not at any specific product, and not with any example output. It sits alongside our broader look at AI ethics and regulation as one concrete, fast-moving case study in how platform economics shift when a new law changes who's liable.

Why this became a business problem, not just an ethics one

For most of the generative-AI boom, non-consensual sexual deepfakes were treated as a content-moderation problem — something platforms handled through takedown policies. That framing broke down for two reasons. First, the technology got good enough and cheap enough that "undressing" a photo no longer requires any technical skill, turning a niche misuse into a mass-market app category. Second, a string of high-profile incidents forced lawmakers' hands: in late January 2024, sexually explicit AI-generated images of Taylor Swift spread virally on X, with one post reportedly viewed 27 million times in 19 hours before the account was suspended — prompting X to briefly block searches for her name and drawing public condemnation from SAG-AFTRA and the White House press secretary. That incident is widely credited with accelerating the federal legislation that followed a year later.

The result: companies operating anywhere near AI image or video generation now face a genuinely new category of legal exposure — not "did we moderate content well," but "did we build something that a new law makes illegal to create, host, or distribute."

The US federal law: TAKE IT DOWN Act

The clearest new liability is federal. President Trump signed the TAKE IT DOWN Act into law on May 19, 2025. It criminalizes knowingly publishing non-consensual intimate imagery — including AI-generated and deepfake images — of both minors and non-consenting adults, and separately requires covered platforms to remove reported non-consensual imagery within a defined window after notice. The FTC's own enforcement guidance confirms enforcement authority became active in May 2026, and Skadden's analysis for businesses frames it explicitly as a platform-compliance obligation, not just a criminal statute aimed at individual bad actors.

The practical shift: a platform that previously could treat non-consensual deepfake removal as a discretionary trust-and-safety choice now has a statutory clock to beat.

State law: Minnesota's first-in-the-nation app ban

States moved faster and further than Washington on one specific front — banning the apps themselves, not just the imagery they produce. California's governor signed three related bills on September 19, 2024: SB 926, creating a new criminal offense for producing sexually explicit deepfakes of real people intended to cause serious emotional distress; SB 981, requiring social platforms to build a reporting mechanism for such content with mandatory temporary blocking pending investigation; and SB 942, a broader AI-content watermarking and provenance-disclosure requirement. All three are described directly on the Governor of California's official announcement.

Minnesota went further still, becoming the first US state to ban nudification apps as a product category, not just the resulting imagery. The bill passed the state Senate 65-0, lets victims sue app operators directly, and empowers the state attorney general to fine violators. It drew a direct legal challenge: xAI sought a federal injunction to block the law from taking effect, and a federal judge denied the emergency request, allowing the statute to stand. For a generative-AI company, that's the sharpest signal yet available: state legislatures are now willing to ban a category of AI tool outright, and courts are, so far, letting them.

The UK closed the "creation" loophole

The US and California approach focuses heavily on distribution and hosting. The UK went one step further. Its Online Safety Act already made it a criminal offense to share or threaten to share intimate images without consent, including deepfakes — but sharing was the trigger, not creation. The Data (Use and Access) Act 2025 closed that gap. As of its commencement on February 6, 2026, it is a criminal offense in the UK to create a non-consensual sexual deepfake of an adult, and — separately and more strictly — to use a nudification tool to create a sexualized image of a child (real or entirely fictional), or to request that someone else do so on your behalf, per legal analysis of the new offenses. The Act also criminalizes supplying nudification tools within the UK, which is the provision with the most direct implication for any company distributing this class of app.

This matters for scope: a US-based generative-AI company with UK users is now potentially exposed to UK criminal law for functionality that may be fully legal to build and offer domestically. Jurisdiction is no longer a formality here.

The EU took a different path: disclosure, not prohibition

The EU AI Act does not ban deepfake generation. Article 50 instead imposes a transparency obligation: deployers of systems that generate or manipulate image, audio, or video content that would appear to be authentic must clearly disclose that it is artificially generated, in a way a person can recognize without needing a separate detection tool. According to the European Commission's own guidance and confirmed by independent legal tracking, these transparency rules took effect August 2, 2026.

The distinction is a real design choice for anyone building compliance into a product: the US and UK approaches are increasingly prohibition-based for this specific use case, while the EU's is disclosure-based — a labeling requirement layered on top of, not instead of, whatever national laws (like the UK's) separately criminalize creation.

App-store enforcement is real but inconsistent

Both major mobile platforms already prohibit this category on paper. Google Play's developer policy bars apps that "degrade or objectify people," including apps that claim to see through clothing or generate nude images; Apple's App Store guidelines separately prohibit overtly sexual content. In practice, enforcement has lagged the policy. In July 2026, San Francisco City Attorney David Chiu sent cease-and-desist letters to both Apple and Google, naming 13 face-swap and nudify apps still listed across their stores despite the stated bans. Apple responded by removing the flagged apps and moving to terminate the associated developer accounts; Google said it suspended all five apps named on its platform.

The gap between written policy and actual enforcement is the operative fact for anyone assessing platform risk: a stated app-store ban is not the same thing as an effective one, and regulators — not just users flagging content — are now the ones forcing the gap closed.

What this means for generative AI businesses

Pull the threads together and a pattern emerges that applies well beyond any single app category:

• Liability is shifting from distributor to builder. Laws like Minnesota's target the tool itself, not just what a user does with it — a materially different risk than traditional platform-moderation liability.
• Jurisdiction is no longer optional to think about. A product legal in its home country can create criminal exposure the moment it has users in the UK, given the creation-based (not just distribution-based) offense.
• "We have a policy against this" is not a defense anymore. Apple and Google both had explicit rules; regulators intervened anyway once enforcement visibly lagged the policy on paper.
• Disclosure and provenance tooling is becoming table stakes, not optional polish. The EU's transparency mandate, plus California's SB 942 watermarking requirement, point toward content-provenance infrastructure (visible or embedded AI-content labeling) becoming a baseline compliance cost for any image/video generation product, adult-content-adjacent or not.

Limitations of this analysis

This piece covers US federal law, California and Minnesota specifically (not all 50 states — other states, including Pennsylvania, have proposed but not yet passed similar bills), UK law, and the EU AI Act's transparency provisions — not every jurisdiction with relevant rules. Enforcement data is thin: beyond the San Francisco cease-and-desist action, there is little public reporting on how consistently these laws are actually being enforced against companies, as opposed to individuals. We also could not verify a credible, named-source estimate of the adult-content-specific AI generation market's size — broader AI image-generation market figures exist from firms like Grand View Research, but a narrower and reliable adult-content submarket figure does not appear to exist publicly, so none is cited here. Finally, this is a fast-moving legal area; several of the sourced provisions (the UK Act's February 2026 commencement, the EU Act's August 2026 transparency rules) took effect within the last several months of this article's publication and may see early enforcement actions or legal challenges that shift the picture further.

This is one entry point into a broader shift already underway across the AI hub — the same compliance and provenance questions are reshaping how companies think about AI model commoditization and the wider AI ethics and regulation landscape more broadly.

The bottom line

The regulatory floor under AI-generated sexual imagery moved substantially in 2026: a live federal law with FTC enforcement, a first-in-the-nation state ban on the tools themselves that has already survived a legal challenge, and a UK law that criminalizes creation rather than just distribution. None of this eliminates the underlying technology risk, but it does change the calculus for anyone building generative AI image or video tools — the question is no longer only "is this content policy compliant," but "which of the jurisdictions we operate in has made this specific capability, or supplying it, a criminal offense." That question didn't exist in this form eighteen months ago. It does now.