Meta's advertising system shifts the cost of moderation failures onto children

Meta cannot explain away the repeated approval of ads promoting child sexual abuse material as occasional failures caused by criminals changing their tactics. The company controls the advertising moderation, distribution, charging, and reporting mechanisms on Facebook and Instagram. When those mechanisms continue to let offending ads through even after government orders, media investigations, and warnings from advocacy groups, automated systems are not a substitute for responsibility. They are exactly what Meta must be held accountable for.

According to a BBC report, the US advocacy group Tech Transparency Project (TTP) said that over the past 10 months Facebook and Instagram had run 332 paid ads containing AI-generated child sexual abuse imagery, 274 of which appeared in August of this year. TTP said 84 of the ads ran in India, and 78 of those were published after the Indian government ordered Meta to remove all ads and content promoting child sexual abuse material and held meetings with senior Meta executives.

These figures come primarily from TTP's investigation and have not been independently verified by the BBC. TTP also described one ad run in India that used AI to manipulate a photograph of a pre-adolescent girl, but the BBC made clear that it had not seen that ad. This evidentiary limitation cannot be ignored, yet it does not erase the questions Meta needs to answer: why, after regulators intervened, did ads allegedly promoting child sexual abuse material still enter its paid distribution system?

The BBC explained that ads on Meta's platforms must first be approved by moderation technology before they can run. The system relies mainly on automation and checks images, video, text, audio, target audiences, and the destinations of links. Meta is therefore not a passive bulletin board displaying third-party information. It decides who can buy reach, which users that content can reach, and it draws revenue from this infrastructure. The harm caused by moderation failures cannot be diluted into the abstraction of 'technical imperfection.'

TTP's testing of Meta's reporting mechanisms further exposed the limits of this system. The group said that of the 84 ads run in India, it was only able to report 55 while the ads were still online. Meta's response to 43 of those was: 'We reviewed the ad and found that it did not violate our advertising standards.' Another 11 were told they had been removed, and one had not received a response. TTP said the reported ads were only removed after it contacted Meta.

The BBC previously encountered similar situations in its reporting. The BBC said that after it reported an ad involving child sexual abuse material through Instagram's reporting system, the platform did not take it down. Only after the BBC approached Meta for comment did the company say it had disabled multiple ads, suspended the related accounts, and additionally removed ads, banned accounts, and blocked URLs.

Journalists and advocacy groups must discover, preserve, and report each ad one by one while the ad is still online, while Meta can let its automated system approve the placement first and then use after-the-fact cleanup as proof that its defenses are effective. This arrangement hands the labor of spotting risk to outside parties, leaves the consequences of dissemination to children and their families, and lets the platform keep control of the rules, the data, and the pace of action.

Meta told the BBC that it has built 'robust defenses' including blocking violating ads and ongoing monitoring, that many flagged ads had already been removed, that most ads received fewer than 200 impressions, and that total spending was under $5,000. But impression counts and ad spending can only describe the scale of dissemination; they cannot prove that the moderation system has fulfilled its duty of protection. For the risk that real children's photos are appropriated and sexualized, a smaller number of impressions is not an exemption from responsibility.

The commercial chain makes it harder for Meta to attribute the problem to outside criminals. TTP said that the overwhelming majority of the 332 ads directed users to AI image or video generation apps, most of which came from Chinese developers. TTP also alleged that some ads were placed by Meta's advertising reseller partners in China; the BBC said these partners, known as 'resellers,' funnel tens of billions of dollars' worth of Chinese advertising into Facebook and Instagram each year. The relevant partners did not respond to TTP's questions.

Meta told the BBC: 'We do not tolerate undressing apps or any form of child exploitation, whether real or AI-generated. These criminals constantly change their methods to evade detection, so we continue to strengthen our detection and enforcement.' The company also said that when it identifies clear child exploitation content, it reports it to the US National Center for Missing and Exploited Children in line with its US legal obligations.

Bhuwan Ribhu, founder of the Indian children's rights network Just Rights for Children, said reporting only to a US body is not adequate. He told the BBC: 'Social media companies are clearly not reporting child sexual abuse material to Indian law enforcement as required by our child protection laws, in brazen violation of Indian law.' The BBC reported that this network of organizations has filed a petition with India's Supreme Court asking it to direct social platforms to mandatorily report such content and identify those promoting it. Multiple Indian agencies also said after the BBC investigation that they were reviewing the matter.

The fact that dissemination routes cross platforms does not lessen the responsibility of the first paid entry point. The BBC previously found that some Meta ads directed users to Telegram, where the material in question was sold for as little as 99 rupees. Shikha Goel, director of the Telangana state cyber security bureau, told the BBC that one of the major challenges in handling child sexual abuse material is that users can easily move from one platform to another, making it harder for police to catch them. Telegram told the BBC at the time that it uses a combination of automated and human moderation and has 'nearly eliminated the public spread of child sexual abuse material.'

Cross-platform movement does indeed make enforcement harder, but Facebook and Instagram provide a buyable, targetable, pre-approved entry point for dissemination. Meta has the power to stop ads from going live, and it also decides when reports are effective, when violating accounts are disabled, and how partners access the ad market. As long as this system keeps leaving pre-launch moderation failures to be borne by children, and then measures effectiveness in low impression counts and post-hoc removals, Meta's so-called defenses will continue to serve the continuous operation of the advertising system rather than putting child safety above its commercial infrastructure.