CNAF anti-fraud algorithm challenged as discriminatory
Fifteen civil society associations filed a case at France's Conseil d'État challenging a CNAF algorithm used since 2011 to detect fraud and overpayments among family allowance recipients. The associations alleged that the algorithm discriminates against vulnerable beneficiaries by assigning higher suspicion scores based on low income, unemployment, and receipt of minimum social benefits, resulting in disproportionate auditing of people in financial difficulty.
What happened
Fifteen civil society associations announced on 16 October 2024 that they had filed a case with France's Conseil d'État to challenge an algorithm used by the Caisse nationale des allocations familiales (CNAF), which has been in use since 2011 to identify family allowance recipients most likely to make errors in their declarations.
The associations gained access to the algorithm's source code from a version used between 2014 and 2018 and stated that it assigns each beneficiary a suspicion score that determines whether they will be audited, with higher scores increasing the probability of control.
Reported impact
The associations argue that the algorithm discriminates against vulnerable beneficiaries by treating low income, unemployment, and receipt of minimum social benefits as factors that increase suspicion scores, resulting in over-control of people in difficulty compared to the rest of the population.
The algorithm analyzes personal data from over 32 million people living in households receiving CAF assistance.
Index analysis
The associations contend that the algorithm's scope of surveillance and its discriminatory effects against vulnerable recipients violate applicable data protection and fairness principles.
Governance controls
The associations requested in July 2024 that the CNAF cease using the algorithm but reported not receiving a response within two months, prompting the judicial recourse to the Conseil d'État.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.