How Meta Ads Let Hundreds of AI Nudify Abuse Ads Through

Introduction

Generative AI has made it easier to create realistic sexualized images and videos, while the distribution of those services increasingly relies on ordinary social platforms. An investigation by the Tech Transparency Project (TTP), published with related reporting from Wired, found hundreds of Facebook and Instagram ads promoting so-called “nudify” apps. Some ads used public photos of real children as source material for explicit videos.

Key findings

• TTP says it identified 332 ads containing child sexual abuse material this year. The vast majority promoted AI apps made in or linked to China.
• Investigators matched several people shown in the ads to real children whose photos had appeared online, including a young member of a European royal family, a 14-year-old Instagram user, and a girl shown in stock photography.
• The ad copy used promises such as “there are no restrictions” and claims that the characters were real. Wired also found hundreds of related ads depicting adults, potentially without their consent.
• After TTP reported more than 300 ads, Meta quickly removed roughly half from its ad library. Investigators said some entries were not initially marked as violating child-exploitation rules, while other ads remained online for days after being reported.
• TTP said the ads collectively reached more than 29,000 people in the European Union and 6,800 in the United Kingdom. About 80 percent were served in the United States. Meta said the flagged ads averaged fewer than 200 impressions and spent less than $5,000 in total, but TTP argued that the platform may be undercounting the problem.

Why it matters

The central issue is not only whether nudify applications are abusive, but whether Meta can stop clearly dangerous advertising before it reaches users. Meta said it does not tolerate nudify apps or child exploitation, and suggested that some material evaded detection because clips were brief or blurred. TTP nevertheless found inconsistent decisions involving identical or near-identical creative, as well as reports that took at least a week to review.

The case highlights three weaknesses. Publicly available images can be repurposed into harmful synthetic material within minutes. Automated detection, human review, and advertiser verification may fail to work as a single control system. And a platform that earns money from advertising faces an obvious incentive problem when high-risk campaigns are not blocked quickly.

TTP also linked some campaigns to Chinese ad-reseller partners and pages that appeared fraudulent or inauthentic. US lawmakers, state attorneys general, and Australia’s eSafety regulator have sought further answers. As regulators increasingly treat AI-generated child sexual abuse material as seriously as conventional material, platforms will need to demonstrate more than post-publication takedowns: they will face pressure to improve pre-launch screening, repeat-ad detection, evidence preservation, and timely reporting.

Source: Ars Technica AI