The campaign the library shows
A single account ran the campaign that the Tech Transparency Project found, promoting Kromix, an app that generates images with AI. Meta's ad library shows that account published 32 ads, live for between 5 and 46 hours, most of them drawing 10 impressions or fewer. All were aimed only at male users. The page behind them had been created on 3 August and had no followers.[1]
The company's published documentation says every ad is checked before it goes live, primarily by automated tools, and that the check runs before an ad starts. The ads went out anyway, and they came down after Wired asked. That sequence supports two readings: the review may never have caught the campaign, or the short run times and low impression counts may show automated systems closing in on the account. The library does not settle which, because it does not say why any individual ad ended.[1]
Which number measures the review?
Meta offers its own figure here: it reports removing 344,000 ads promoting such services between November and January, and suing at least one firm that markets them. Cindy Southworth, who leads women's safety at the company, says the space is highly adversarial and that its defences are strengthened accordingly.[1]
A removal total counts the ads that left the platform after publication. The output of the review that runs before publication — how many ads it stopped at the gate — sits outside that total. A rising figure therefore does not establish that the gate is getting stronger; the same rise is consistent with more ads getting past it. The number that would separate those two possibilities has not been published.[1]
What would close the gap?
Two features of the account were visible in the library: the page had been created on 3 August and had no followers. Whether the review weighs page age and audience alongside the creative has not been disclosed. The classifier may instead work mainly from the imagery and the voice-over, in which case those account features would not have changed the outcome. Only publishing the review's criteria resolves that.[1]
The same pattern shows on Apple's side: the app was pulled from the App Store after Wired's inquiry, and San Francisco's city attorney had sent Apple a cease-and-desist earlier this summer naming 13 apps. In both cases the remedy arrived through an outside question rather than the system's own detection. That yields a testable expectation: if Meta's next transparency report places a rejected-at-review count beside its removal total, the gate's performance becomes measurable; if it publishes removals alone, the next campaign will again be ended by whoever asks first.[1]