Tax expenditures, subsidies and fraud, Bruno Retailleau
Tax and social security fraud: €9bn expected from fighting fraud: 5 tax and benefits fraud, 4 social
Direction of change: increase, compared with current law.
The lever. Combating tax, contribution and benefit fraud: inspection staff numbers, sanctions, expected revenue. Direction rule: ↑ stronger action.
« Combien, et d’où vient le chiffre. 4 Md€ sur la sphère sociale ; avec les 5 Md€ de fraude fiscale et aux aides publiques (fiche 5), 9 Md€ »1Translation by the editorsHow much, and where the figure comes from. €4bn in the social sphere; with the €5bn of tax fraud and fraud on public aid (fact sheet 5), €9bn
Encoding
No cost-schedule parameter for this position: see the costing for the reason.
Costing
Announced by the candidate
+€9bn
per year, gain for the public accounts. €9bn in 2032: €5bn from tax fraud and fraud involving public aid (administrative freezing of assets, AI-targeted checks, secured aid; press fact sheet no. 5, measure no. 120) and €4bn from social fraud (fact sheet no. 10)
Published costings
none
No external costing of this position recorded to date.
Our calculation
not costed
Revenue and savings target whose tools are described (administrative freezing of assets, AI targeting of checks, cross-matching of files, offsetting of debts against benefits), with no staffing or audit budget: the cost schedule costs revenue recovered on the basis of the resources committed, not a target.
Current law, sources and limits of the cost schedule
Current law. Tax audits in 2024: €20.1bn of duties and penalties assessed for collection, €11.4bn actually collected (Cour des comptes, December 2025). No official costing of tax fraud: for the Cour, « nul ne peut aujourd'hui se hasarder à affirmer que ce montant est proche de 30 ou de 130 Md€ » [“no one today can venture to assert that this amount is close to €30bn or €130bn”]. VAT gap: €6bn to €10bn (DGFiP). Social security fraud estimated at €13bn, of which €2.1bn detected and €0.6bn recovered (HCFiPS, 2024). Anti-fraud bill under discussion in 2026.
Horizon. Phased in over several years; full year 2032 to be justified by resources (staff, tools).
Limits. The amounts of « fraude totale » [“total fraud”] often cited (€80bn to €100bn, Solidaires) are judged by the Cour des comptes to have no method. A promise of yield from the fight against fraud is costed on the revenue recovered, not on the estimated fraud, and related to the resources committed. Coefficient calculated by the editors (10% of the €11.4bn recovered; €20.1bn issued for collection as the upper bound). The costing is fragile: low confidence index.
- Cour des comptes, La lutte contre la fraude fiscale, 2025-12-16: « Celles-ci sont passées de 12,2 Md€ en 2015 à 11,4 Md€ en 2024 ... ces résultats retrouvent leur niveau de départ (20,1 Md€ en 2024 contre 21,2 Md€ en 2015) ... Nul ne peut aujourd'hui se hasarder à affirmer que ce montant est proche de 30 ou de 130 Md€ ... La fourchette finalement dégagée par la DGFiP (entre 6 et 10 Md€ ...) »
- HCFiPS, Fraude sociale 2024, infographie, 2024-09-25: « MONTANTS ÉVALUÉS ... 13 MD€ ; FRAUDES CONSTATÉES ET STOPPÉES 2,1 MD€ DONT 0,5 MD€ ÉVITÉES ; MONTANTS RECOUVRÉS 0,6 MD€ »
How to read these amounts. These are static costs and revenues, with no feedback effect, over a full year: nothing on growth, employment or the revenue that economic activity would gain or lose. A higher net cost does not on its own make a worse policy, nor net savings a better one: each programme rests on an economic model (demand-led stimulus, lower compulsory levies, consolidation of the public accounts…) that the cost schedule does not settle. GDP itself measures a volume of activity, not well-being or the usefulness of what is produced. Economic models and feedback effects.
- avecretailleau.fr, fiches presse « Rétablir l'ordre dans les comptes publics » (12 octobre 2026), fiche n° 10, p. 12. Open the document at this page.
Contest this encoding (prefilled email; procedure and deadlines)