For several years now, the tools used for tax audits have been based on increased recourse to the exploitation of mass data, or data mining, through artificial intelligence. The General Directorate of Public Finances (DGFiP) can thus identify fraud profiles by analyzing and cross-checking, using algorithms, all the information it has, and establish lists of taxpayers to be controlled.
Originally, these cross-checks could only relate to data from files of the tax administration and other French or foreign administrations, or from private databases.
In 2021, an additional step was taken with the technique of “web scraping”. Implemented on an experimental basis for a period of three years, this technique allows the tax administration to collect and use – through computerized and automated processing – personal data published on social networks (Facebook, Twitter, Instagram, LinkedIn, etc.) and on online sales sites like Leboncoin, Vinted, eBay, etc.
Internet survey
The only condition for being able to “swallow”: this data must be accessible to everyone, without it being necessary to be registered on the site or enter a password to access it. In addition, the collection and processing of this data can only be carried out to pursue one of the following three objectives: detect “false” tax domiciliations abroad, identify undeclared professional activities and hidden smuggling activities. and counterfeiting.
Judging the “promising results” – the results of this experiment have, for the moment, not given rise to any report, according to the Solidaires public finances union –, the finance bill for 2024 plans to extend it for two more years. In doing so, it broadens its scope, both in terms of data collected and breaches sought, and authorizes “tax agents to carry out pseudonymous investigations on websites, social networks and messaging applications”.
Combining AI and aerial views
Another flagship project which has largely mobilized artificial intelligence, and on which Bercy has communicated extensively, is the system called “Innovative Land”. It is based on the use of aerial images from the National Institute of Geographic and Forestry Information. The algorithms implemented by the DGFiP services made it possible to extract the contours of built buildings, particularly swimming pools.
You have 31.86% of this article left to read. The rest is reserved for subscribers.