- Ad Referrals: Meta has deployed new AI tools to detect ads that lead people to child sexual abuse material outside its platforms.
- Link Blocking: Meta blocks violating links and removes other ads, posts and comments carrying them.
- Account Returns: Meta says stronger detection identifies people trying to create new accounts after removal.
- Enforcement Scale: Meta reports acting on 33.2 million pieces of child sexual exploitation content in January–June 2026, without specifying the new tools’ contribution.
- Removal Delays: In August, reported ads kept reaching users before removal, the Tech Transparency Project watchdog found.
Meta says it has deployed new AI tools to detect Facebook and Instagram ads that direct people to child sexual abuse material elsewhere online. The company is strengthening checks on where ads send users, aiming to interrupt referrals that harmless-looking advertising can conceal.
Meta said on October 7 that it disabled related violating accounts and blocked external links after investigating the advertising scheme. It then deployed additional detection measures. The company has not disclosed the deployment dates or a test measuring how much the new tools reduce exposure.
Following Ads to Their Destinations
One new detector uses a large language model to flag what Meta calls signposting: apparently benign ad content suspected of directing people to prohibited activity outside its services. Meta says improved destination assessment helps it identify violating links and act against the accounts distributing them.
Once it blocks a link, Meta says it searches for other ads, posts and comments carrying that address and removes them. Its systems also reject ads containing blocked links when advertisers upload them and prevent those links from being posted on Facebook, Instagram and its Threads social network.
Additional AI scans seek child exploitation content missed by earlier systems. Stronger repeat-offender detection looks for people creating new accounts after removal. A red-teaming AI agent tests Meta’s own defenses for weaknesses, with the stated aim of finding abusive tactics before they spread.
In July, Meta explained how it uses AI detection of suspicious off-platform links combined with other child-exploitation signals, advertiser investigations and URL blocking.
Methods similar to Microsoft’s PhotoDNA compare digital fingerprints of images to identify copies of already-known material. Meta says it also uses image and video matching, including in external-link previews. Suspected referrals require examining the relationship between an ad, its destination and the account promoting it, even when the visible ad contains no illegal image.
What the Enforcement Totals Measure
Meta reports taking action against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram worldwide in January–June 2026. More than 97% of that actioned content was found before users reported it.
So far, Meta has not published an ad-only total or a performance breakdown for these tools.
The Tech Transparency Project, an investigative watchdog, identified 332 ads containing child sexual abuse imagery that had passed Meta’s review. Its September investigation used Meta’s Ad Library, a searchable advertising archive. The archive’s regional retention rules gave researchers uneven access to past ads.
TTP recorded more than 29,000 people reached in the European Union and 6,800 in the United Kingdom. Audience figures for other regions were unavailable in the data it could examine.
The watchdog had shared an initial group of ads with WIRED in early August. Meta said many of those ads predated an AI system intended to block violating ads during upload. TTP subsequently identified hundreds more ads running after that initial report.
In one instance, the watchdog reported an ad on August 14 after it had reached four users in the European Union. Another 330 EU users saw it before Meta removed it days later. WIRED reviewed screenshots showing additional views during removal delays for its September 8 report.
The watchdog said 183 of its 332 identified ads had been removed by September 1, with the remaining 149 removed after it shared its findings on September 2. It also reported that 113 removal notices initially used adult-content or generic policy labels before being changed to child-exploitation labels after questions. The watchdog questioned the reliability of Meta’s self-reported enforcement statistics.
Meta spokesperson Tracy Clayton told WIRED in September that many identified ads had already been removed, most had fewer than 200 impressions and payments totaled less than $5,000. Meta also said brief or obscured appearances of children made detection harder.
TTP director Katie Paul told WIRED that new ads had appeared since the watchdog’s September 8 report, some featuring children whose images it had already sent to Meta.
San Francisco City Attorney David Chiu said his office independently reviewed many of the identified ads. His September 9 letter to Meta sought explanations of review failures, reporting and repeat offenders, and proposed public evaluation of the systems’ effectiveness. Meta disputed the city’s jurisdiction, saying there was no indication the ads ran in San Francisco; the city attorney’s office replied that local consumers could access them online.
Apps Beyond Advertising
Some of the investigated ads led to AI image and video apps with capabilities for turning photographs of people into sexual imagery, according to TTP. Emma Hardy, communications director at the Internet Watch Foundation, warned that ordinary photographs of children could be turned into abuse material.
Apple told WIRED it had removed most of the apps identified in the research and taken action against the rest, as well as removing additional apps found by the publication. It said developers initially submitted compliant products and later introduced the prohibited image-manipulation capabilities. Google said the apps in the report had been removed from its store. Removing apps limits their distribution through the stores; Meta’s destination and account checks act on the advertising that sends people to them.
