Document tampering fraud

In May 2019, a video of Nancy Pelosi, then Speaker of the House of Representatives, circulated slowed down to make her speech sound slurred: no artificial intelligence was involved, editing software was enough, and the press used the word shallowfake to describe this cheap manipulation, as opposed to a deepfake. In insurance, document tampering fraud means falsifying or fabricating the documents that support a claim. An invoice with a changed amount or date, a damage photo lifted from another claim or from the internet, an altered medical certificate or sick note, an inflated garage estimate, a reassembled bank statement. Consumer editing software was already enough; generative AI has brought the cost down, since a picture of a dented bumper or an appliance invoice bearing a retailer's letterhead now takes a few seconds and no skill. Deepfakes, for their part, cover the synthetic video and voice used to impersonate a person, and a synthetic identity fabricates an entire customer. Here, the person and often the loss are real, only the document lies. The countermeasures rank by cost. A photo's EXIF metadata (date, device, geolocation) gives away an image taken before the loss or downloaded, but it can be stripped or edited with no effort. Reverse image search finds a photo already published online. Forensic analysis looks for recompression traces, inconsistent shadows, a font that changes halfway through a line. Some insurers also require photos to be taken inside their app, time-stamped and geolocated at the moment of the inspection, and the C2PA standard offers a way to attach signed provenance to files. The safest countermeasure is still the oldest: verify at the source, by calling the shop, the garage or the hospital that supposedly issued the document.