Meta Runs Child Abuse Ads in India
It is proposed, with the seriousness the subject deserves, that Meta formalise what its own trust-and-safety architecture already practises, and cease pretending that the difference between a violation prevented and a violation merely discovered is a difference of principle rather than of timing.
The BBC Eye investigation into advertisements promoting child sexual abuse material running on Meta’s platforms in India did not, if the report is credited, uncover a system that fails occasionally under unforeseeable pressure. It uncovered a system operating precisely as designed: content reviewed at scale, flagged at scale, removed at scale, and - inevitably, given the mathematics of scale - permitted to run at scale in the interval between posting and detection. Meta’s own public position, that it invests heavily in safety tooling and acts on reports when they arrive, is not a rebuttal of this. It is a confession of the operating principle. The ad ran. It was reported. It was, presumably, eventually removed. Three verbs, in that order, describe a tolerance, not a prohibition.
Given this, the reasonable administrator does not ask how the tolerance can be eliminated - the company’s own quarterly transparency reports suggest it regards elimination as a rounding problem, not a design flaw - but how the tolerance, being real, might be managed honestly rather than discovered periodically by journalists. It is proposed that Meta publish, for each major advertising market including India, a Residual Harm Allowance: a stated, quarterly figure representing the volume of exploitative advertising the company’s own detection latency permits to circulate before removal. This is not a call for the company to permit more harm. It is a call for the company to stop describing an already-existing permission as an accident.
The advantages of candour here are considerable. An allowance that is published can be audited; a failure that is merely apologised for, market by market, investigation by investigation, cannot. Regulators in Delhi, currently limited to reacting to BBC Eye’s findings after the fact, would instead be handed a number against which to measure compliance - the same actuarial courtesy insurers extend to flood zones and airlines extend to runway incursions. Meta’s shareholders, who presently absorb reputational shock as an unbudgeted cost each time an investigation lands, would instead see the cost priced in advance, the way any well-run enterprise prices in the cost of its own known defects. And the advertising ecosystem that currently treats such content as an anomaly requiring apology would instead treat it as a line item requiring a rate card - which, functionally, is what it has been all along, the only novelty being that someone would finally write the number down.
One anticipates the objection that no acceptable number for a child’s exploitation being auctioned as ad inventory can exist, that the arithmetic is obscene before the first digit is entered. This objection is correct, and it is precisely the objection Meta’s engineers overrule every day the system runs as configured rather than as advertised. The ledger changes nothing about the harm. It only removes the comfort of calling a policy an accident. Somewhere in Mumbai or Lucknow tonight, an ad slot is clearing, a bid is winning, a child does not know the auction has her in it - and the report exists, and the report will be answered, and the market will open again tomorrow at the usual hour.