Meta has a rule against ads for apps that strip clothes off women’s photos using AI. And yet, according to a new report, Meta ran roughly 7,600 ads for those kinds of apps on Facebook and Instagram between April and June of this year, delivered through one of its own official Chinese advertising partners. The gap between policy and practice is huge, and something that cannot be ignored. 

The report comes from the Tech Transparency Project, the research arm of watchdog group Campaign for Accountability, and it traces the ads back to GatherOne Inc., a Beijing-based agency that connects Chinese companies to Meta’s ad platforms even though Facebook and Instagram are both blocked inside China. GatherOne is one of just 11 official agents listed on Meta’s China-facing business site, and its CEO has posed for a photo standing next to top Meta executives at a Hong Kong event with Meta’s logo prominently displayed behind them. That’s not a rogue reseller operating in the shadows. That’s a partner with a seat at the table. 

One app in particular, an Android downloader called BAfter, is where this stops being an ad-policy embarrassment and starts getting much darker. TTP found a “ShareZone” section inside the app offering pornographic face-swap content, and multiple Google Play reviewers flagged what they described as AI-generated videos involving minors, despite the app being listed as suitable for all ages. BAfter disappeared from the Play Store on July 21, and GatherOne says it’s permanently cut the developer’s advertising access. Meta says it banned the app and stopped allowing links to several other nudify tools. Cindy Southworth, Meta’s head of women’s safety policy, insists the company doesn’t allow “non-consensual intimate imagery or nudify apps” and takes “aggressive steps to combat them.” But how aggressive are they being if ads to these companies were running until TTP brought them to light? 

Not very, and the reason comes down to money. Meta made $18.4 billion from China last year, roughly 11 percent of its global revenue and more than double what it made there just two years earlier. Reuters reported in December that internal Meta documents put about 19 percent of that China revenue (nearly $3.5 billion) behind ads for banned content categories like pornography, scams, and illegal gambling. GatherOne was named specifically as an agency whose ads Meta’s own testing linked to scam services. A former Meta policy director, Sarah Wynn-Williams, testified to Congress last year that China remains central to the company’s strategy in ways that go well beyond advertising. Whatever the internal reasoning, the outcome is the same. A platform with explicit rules about this exact category of harm kept running ads that violated them, at scale, for months, through a partner Meta itself vouches for. 

Meanwhile, a different social media platform is suing to block a state law that would hold AI companies liable when their tools get used to generate nonconsensual nude images. Elon Musk’s xAI argues that the statute is unconstitutionally broad because it doesn’t require proof the platform intended misuse. Meta’s ad failure is the practical argument against the legal theory. Nudification tools don’t need a company’s blessing to cause damage. They need distribution, and that’s what these platforms provide. A George Mason University survey found more than half of surveyed teens had used nudification tools on themselves or others, and researcher Chad Steel noted teens already know how to reach these apps directly, so app-store or ad-platform crackdowns won’t stop access. So what will? It’s not clear. 

The same industry insisting that broad liability laws limit legitimate free speech is also the industry that, given a policy and a profit incentive, will always choose profit. Congress passed the Take It Down Act last year as an attempt to force platforms to remove nonconsensual sexual content, but the FTC only started enforcing it earlier this year. It’s not clear if this act will actually do anything. Because when a company can write the rule, break the rule, and still argue in court that the rule shouldn’t exist, what can even be done?