The investigation by Bloomberg and the Japan Times traced a pattern across Meta’s platforms: roughly 7,600 ads, funneled through Chinese advertisers, promoting apps that generate nude images from ordinary photos of people — including, in the ads’ own marketing, photos of strangers. The apps’ pitch was explicit. Upload a picture, and the software removes the clothing.
Meta said it has removed the offending ads and taken action against the accounts behind them. The scale is the problem. Seven thousand six hundred ads is not a slip or a single bad actor; it is a pipeline, and the fact that it ran on Facebook and Instagram for as long as it did is, for critics of the company’s moderation systems, evidence that its automated review is not up to the task.
The apps belong to a category that has become one of the most harmful corners of the AI boom: tools that produce nonconsensual intimate images. The technology is simple enough that small developers build and market these apps at scale, and the images they produce are used for harassment, extortion and humiliation. Lawmakers in several countries have been drafting laws to criminalize the tools themselves; enforcement so far has lagged the spread.
How the ads slipped through is a question about Meta’s ad review architecture. The company reviews a fraction of ads by machine, and its automated systems are trained to catch obvious violations — nudity, scams, prohibited products. AI undressing apps present a subtler problem: the ads themselves may show clothed people and carefully worded copy, with the app’s actual function described in the landing page or the app store listing. An ad that passes the text and image checks can still send users to something deeply illegal in most jurisdictions.
Meta’s response has been the standard one: the ads are removed, the accounts are banned, the company says it will improve its detection. The pattern is familiar from years of content moderation battles — the same rhythm of investigation, takedown, and promise of better systems. What has changed is the technology: the apps being advertised have gotten good enough that the harm is no longer theoretical, and the ads have gotten sophisticated enough to evade the filters.
The stock market added its own verdict. Meta shares have fallen for seven consecutive sessions, a slide of about 13 percent, as investors weigh the moderation failures against the company’s heavy spending on AI infrastructure. The connection between the ad scandal and the stock decline is not direct — the slide has broader causes — but the timing compounds the pressure on a company that has bet its future on being trusted with more of users’ lives.
The episode also exposes the jurisdictional gap in AI enforcement. The advertisers are Chinese; the platforms are American; the victims are global. Meta can block the ads it finds, but it cannot shut down the app developers, and it has limited ability to pursue them across borders. The apps are sold through app stores and websites that Meta does not control, and the same developer can return under a new name the day after a ban.
Industry observers say the problem will only get harder. Image generation is improving faster than detection, and the same models that create harmless content can be repurposed. The technical answer — watermarking, provenance systems, better classifiers — is real but incomplete, because the people building these tools are not playing by the same rules.
For Meta, the stakes go beyond reputation. The company is fighting legislative battles in the US and Europe over online safety, and each scandal gives regulators a concrete exhibit for why platforms cannot be left to police themselves. The 7,600 ads are, in that context, more than a moderation failure; they are a data point in the argument that the platforms’ incentives do not align with the public’s.
The investigation also detailed how the ad pipeline worked. The ads were purchased through Chinese advertising accounts, often in batches, with landing pages that rotated to evade detection — a structure familiar from the counterfeit and scam ads that Meta has fought for years. What is new is the product: AI undressing apps did not exist at scale when Meta built its current ad review systems, and the company has been retrofitting its classifiers to catch a category that did not exist when they were trained.
Meta’s statement promised improved detection and swifter action. The company has made such promises before, and its systems have genuinely improved over the years — the volume of ads that slip through today is a fraction of what it was a decade ago. But the bar for this category is absolute: a single undressing app ad reaching a single victim is one too many, and 7,600 of them is a number no remediation statement can make disappear. The question now is whether Meta treats this as another incident to manage or as a signal that its moderation architecture needs a deeper fix.
