San Francisco Orders Meta to Stop ‘Allowing’ AI Child Abuse Ads

This is not a gray-area content-moderation spat. San Francisco hit Meta with a cease-and-desist after researchers documented hundreds of paid ads that transformed stills of minors—including confirmed real children—into short clips that can depict sexual acts, then funneled users toward apps that can undress people or generate nonconsensual intimate imagery.

WIRED’s reporting and the Tech Transparency Project’s work show the problem persisted after earlier alerts. When a platform that claims every ad is reviewed still accepts money to distribute AI child sexual abuse marketing, the failure is commercial and operational, not merely a stray upload.

We’re not hedging here. Useful AI image tools are not the villain in this story; Meta’s ad machine failing to keep AI-generated child sexual abuse out of paid inventory is. City Attorney David Chiu’s letter says Meta failed to address a known issue while profiting from it, and that the scale and persistence look like systemic weakness, not a one-off miss.

The numbers in the reporting are damning enough without embroidery: more than 350 ads in recent months, more than 250 of them after WIRED’s early-August batch, reach into the EU, US, Australia, and India. TTP says some reports sat more than a week while ads kept accruing views. After the September 8 TTP report, Katie Paul said brand-new ads featuring some of the same children still appeared.

Meta’s posture—ads removed, low impression counts for many, total spend under $5,000 for the researched set, and a jurisdictional shrug that San Francisco may not be where ads showed—misses the moral and practical point. A cheap campaign can still be catastrophic for the children depicted. Arguing venue while researchers keep finding nudify and abuse ads is the wrong reflex.

If this trajectory becomes normal, AI makes child exploitation advertising easier to spin up, mutate, and reupload, while platforms treat takedowns as the product. Enforcement has to move upstream: advertiser identity, repeat-offender kills, NCMEC escalation that actually fires, and ad-review systems that catch AI CSAM before payment clears. Eyes open means saying generative tools amplify harm when distribution incentives are sloppy.

Meta’s policies already ban child sexual exploitation, adult exploitation, and promotion of nudify apps, and the company says ads get automated and sometimes manual review. It has removed large volumes of nudify ads and pursued at least one related lawsuit. The city wants answers within 28 days on detection failures, escalation, and advertiser accountability.