CNAF's Algorithme de Notation — France's Decade-Old Welfare Risk-Score Facing a 25-Organisation Legal Challenge

Since 2010, France's national family-benefits agency, the Caisse nationale des allocations familiales (CNAF), has run a machine-learning risk-scoring system that assigns every person living in a household receiving CNAF benefits a suspicion score between 0 and 1, recalculated monthly, to decide who is investigated for fraud or overpayment. The system began as a local pilot in Bordeaux in the late 2000s and was rolled out nationwide without public consultation.
In a six-month investigation, Lighthouse Reports and Le Monde used freedom-of-information requests and legal appeals to obtain and test the underlying source code. They found the score rises for people with low income, unemployment, a disability allowance (AAH), the RSA minimum-income benefit, single parenthood, or a high rent-to-income ratio, while it is close to impossible for wealthier claimants to score high enough to be flagged. Seven out of every ten people subjected to CNAF's most invasive fraud checks — including home visits where inspectors count toothbrushes to estimate household size, neighbour interviews and bank-record checks — were selected by the algorithm.
On 15 October 2024, Amnesty International and 14 other civil-society and disability-rights organisations filed a complaint with France's Conseil d'État (Council of State) seeking to have the system stopped as discriminatory; ten further organisations joined the case in January 2026. Under this legal pressure, CNAF published the source code of its current algorithm on 15 January 2026, more than a decade after first deploying it.
As of this writing the Conseil d'État case is unresolved. No independent evaluation cited by these sources shows the score actually improved fraud detection rates or reduced overpayment losses relative to prior methods; the documented effect is that it concentrates invasive scrutiny on already-vulnerable claimants.

Read the full analysis: https://www.ldh-france.org/algorithme-discriminatoire-de-notation-de-la-cnaf-10-nouvelles-organisations-se-joignent-a-laffaire-devant-le-conseil-detat/