Linking every fact in a claim to its source doesn't work on a fake document, which makes the checks outside the file the ones that count.
By Mike Cooper ·
Allianz UK shared a claim that shows why. It came with a photo of a van with a cracked front bumper and a repair invoice of around £1,000.
But the accident never happened. Someone at Allianz found the same photo on the van owner's business page, and in that one the bumper was fine. The crack had been added with AI photo editing, and someone else had made the claim in the owner's name.
An AI reading that claim would link the damage to the photo and the amount to the invoice, and every link would hold up. A citation shows where a fact came from, so a handler can check the AI read it right. It can't show whether the document is true, because that answer isn't in the file.
Allianz caught a second fake the same way. Someone staged a crash with a Mercedes and used AI to give it the plate of a car for sale online. Investigators saw it was a different model.
I think this is the part people skip when they ask if AI can be trusted with claims. For the van it means searching for the photo online, calling the garage on the invoice and checking the plate against the car. None of that is in what the customer sent.
Which facts in your last claim file did anyone check against something outside it?