Meta's reviewers cleared 43 of the 55 CSAM ads that TTP managed to report

Meta's flow for reporting an ad only works while the ad is running. The Tech Transparency Project says that gate is why it could file on 55 of the 84 ads it found running in India ^[12], leaving 29 with no route in at all ^[1]. The ceiling on what any outsider can prove is set there: evidence of a paid ad exists for as long as someone keeps paying to serve it.

Of the 55 it did file, 43 came back with "We've taken a look and found that this ad doesn't go against our Advertising Standards" ^[13]. Eleven were reported as already removed and one never got an answer ^[14], which accounts for the whole set ^[5]. On the subset where a reviewer saw the complaint and answered it, the answer was that the ad was fine 78% of the time ^[2].

Meta's July post said anyone on its platforms can report ads they believe violate its policies ^[15]. That is an accurate description of the button. It is not a description of what the button did here. Meta's statement to the BBC puts the problem at detection, saying criminals constantly shift tactics to evade it ^[7]. The 43 replies describe a different failure: creative that was found, surfaced, looked at, and approved. Better classifiers fix the first. Only the policy text and the reviewer instructions fix the second, and more reviewers reading the same guidance will just clear the ads faster.

The monthly numbers can be read as a warning about ad review losing ground. TTP dates 274 of the 332 ads to August, 83% of the total ^[1][3], and that curve invites a reading where generative abuse is simply outrunning ad review. The report as published does not say how TTP searched across the ten months, so the concentration is a find rate, not a measured base rate of ad submissions. What the material does establish is duplication and persistence: TTP says one India ad was posted from more than 50 different handles over a fortnight in August ^[11], and that more than a dozen ads appeared after it had told Meta what it had found ^[6]. Deduplicate on the handle and that is 50 enforcement actions. Deduplicate on the creative or the voiceover and it is one.

The India numbers are the sharpest part. TTP puts 84 ads in India, with 78 of them published after the Indian government's removal order and its meeting with senior Meta officials ^[4], which is 93% of the country total ^[4]. Distribution of CSAM is a criminal offence in India, and Meta's own policy bars ads that sexually exploit or endanger children ^[20]. Bhuwan Ribhu of Just Rights for Children says platforms are flouting Indian law by not reporting this content to Indian law enforcement ^[17]; Meta says it reports apparent child exploitation to NCMEC as its legal obligations require ^[16]. Those two statements can both hold, and the gap between them is what the Supreme Court petition is about ^[18].

The transferable exercise, for anyone who owns a report queue: sort last month's reports on two axes, whether the item was still live when the report arrived, and whether first-pass review cleared it. The cell nobody inspects is live-and-cleared. Pull 50 of those and adjudicate them yourself against your written policy. A large disagreement rate tells you what your reviewers were instructed to approve. A large share of reports that never arrived tells you your report button is measuring flight length.