False documents: AI sows chaos in companies, and no one is ready

The advances in artificial intelligence, in particular in the generation of images, greatly exceed the simple entertainment or the creation of amusing visuals. Another facet emerges as to AI uses and specifically concerns fraud to falsified documents. Professionals in the legal sectors, human resources and accounting are particularly exposed.

AI simplifies the falsification of documents

Modifying a checkout to artificially inflate a restaurant note is relatively simple using tools like Chatgpt. However, most of the applications used to scan these supporting documents and generate costs of costs do not have, to date, safety mechanisms necessary to identify this type of fraud. The same goes to declare fake claims to insurance, especially in the vehicle damage sector.

No sector escapes the wave of falsifications made possible by the generative AI: identity theft, false banking or insurance documents, rigged administrative papers, manipulated accounting assessments, modified tax declarations, counterfeit diplomas … The list lengthens as the tools are perfected.

According to a study conducted by Ipsos Digital, 1 in 5 people indicates that it has been faced with a falsified document in the context of its professional activity. An expansion figure, which is explained … by the democratization of generative artificial intelligence.

Adopt new verification protocols

Therefore, new means of verification are essential. If it is technologically difficult to distinguish the true from the false in a automated way, at a time when the images produced by the AI ​​are still not watermark, a solution would consist in mixing the human with the machine.

For example, humanly check the references on a transmitted CV, cross the sources for the verification of proof of address, manually calculate contributions on a salary slip in order to ensure their consistency …

Faced with these drifts, some experts bet on a technological response: use AI to counter AI. Concretely, models specifically trained in documentary analysis could be able to identify inconsistencies, anomalies or suspicious elements in falsified files. If the promise is real, the cost of these solutions may however slow down adoption, especially among SMEs.