Are the most precarious accused of being more likely to cheat child benefit? The association for the defense of digital freedoms La Quadrature du net published an investigation on Monday in which it denounced an algorithm that it considers discriminatory used by the National Fund for Family Allowances (CNAF).
Used by the CNAF since 2011, this statistical tool aims to identify among the 13.5 million beneficiaries most likely to make errors in their declaration, CNAF director general Nicolas Grivel told AFP.
However, beneficiaries of certain minimum social benefits have to fill out quarterly tax returns with complex forms and are more at risk of making mistakes, says the CNAF.
Read alsoTo reduce unemployment, the government is considering reducing the number of contractual terminations
"Our objective is to limit as much as possible reporting errors and their consequences in terms of generating undue amounts," says Nicolas Grivel, who says that it is "not discriminatory" and does not necessarily target "the poorest people but those whose incomes vary".
But according to La Quadrature du Net, which obtained the computer code for this tool via a request for access to administrative documents, each beneficiary appears in the CNAF files according to a "suspicion score". Updated every first of the month, this score ranges from 0 to 1, and the closer it gets to 1, the more suspicious the recipient is.
'A particularly pernicious surveillance system'
From the outset, this score would be higher for people with low incomes, the unemployed, recipients of minimum social benefits and residents of "disadvantaged" neighbourhoods, and lower for the most affluent.
As a result, the poorest would be more likely to be stopped after a simple change of circumstances, such as a move, than a more affluent beneficiary.
"Little by little, light is being shed on a particularly pernicious mass surveillance system," denounces the association, which has only been able to access the source code of two algorithm models used by the CAF between 2010 and 2018, the body refusing to share the current version of the system.
In the spring, the Changer de Cap collective had already mentioned the "opacity" of these algorithms and called for the implementation of random checks. La Quadrature du net points the finger at other institutions such as the Health Insurance, the Old Age Insurance, the Agricultural Social Mutualities or the Pôle Emploi, which also use automated systems to fight against social assistance fraud.
Very difficult to assess, this would represent a loss of several billion euros for the State. For social benefits alone, it is estimated at between 6 and 8 billion per year, according to the Court of Auditors. Last May, the government unveiled a plan to save money and double recoveries by 2027.