What the ad library fixes and what it leaves open
Researchers at the Tech Transparency Project counted more than 250 further ads carrying child sexual abuse material on Meta's platforms since the start of August, taking the total past 350 since the end of last year. This time images of real children were identified in the ads: four minors were traced, among them two teenagers with public accounts on Meta's own platforms and a stock photo model labelled as pre-teen. The company's policies say it reviews all ads that run on its platforms; the ads ran anyway, and in some instances reviewing a reported live ad took a week.[1]
On 18 August this column argued that the removal total Meta publishes cannot show why the pre-publication review let the ads through, because that total does not count what was stopped at the gate. Since then the company's answer has gained a layer: new AI tools designed to better detect and block harmful ads. The counts rose anyway, and no figure showing what those tools stop at the gate has been published. What exists is a record of removals rather than a record of gate stops.[1], [3]
The same step on a different platform
Moustafa Ayad of the Institute of Strategic Dialogue counted 150 videos across 71 accounts on TikTok, with more than 5.4 million views in total. A TikTok spokesperson called the report's methodology vague and its conclusions poorly evidenced, and said the platform uses a combination of advanced moderation technology, safety teams and partnerships to proactively detect and remove violative content. The same statement said all the content referenced in the report has been removed; dozens of the videos were deleted after they were shared with the platform. Ayad says some of the content survived for months.[2]
In both files the removals are real and documented, but in both the removal began with a list handed in from outside: a group of researchers named the items one by one, and the platform then deleted them. In that order a removal total measures how far reporting reached rather than what the review caught. That is not the only reading: both platforms may be stopping items at the gate and simply not publishing that number, in which case the totals understate detection rather than describe it. Only a published gate count separates the two.[1], [2]
Which figure would settle this?
What would close this is not a complex document but a single broken-out number: how many ads the pre-publication review stopped, reported separately for AI generated sexual content. If Meta publishes that breakdown, an outside reader can set it against the 350 ads the researchers counted and see for themselves whether the gate is working. Without it, only the count of removals remains, and that count grows each time somebody reports.[1]