Does the French State still need 105,000 tax agents in 2026?

Does the French State still need 105,000 tax agents in 2026?

The national diagnosis is devastating, and it is officially documented

• French trust in the State for “the proper use of public funds” fell from 33% to 22% between 2023 and 2025, according to the Compulsory Levies Council barometer by Harris Interactive in November 2025. That is an 11-point drop in two years, unprecedented in scale.
• 78% of French people consider the level of taxation too high, and 75% now believe that expatriation can be justified in order to reduce one’s tax burden.
• 59% of French people say they would accept a reduction in the level of public services in exchange for lighter taxation.
• 27% of young people aged 18 to 25 believe it is justified to cheat on taxes if one has the opportunity.
• 44% of tax fraud cases transmitted to prosecutors in 2024 were dismissed without further action, and only 27% were brought before a court.
• The Court of Auditors itself describes French tax fraud as a phenomenon that is “poorly identified, poorly quantified and poorly handled.”

What we are collectively experiencing is not a simple case of tax fatigue. It is a progressive rupture of consent to tax, officially diagnosed by the Compulsory Levies Council itself, which speaks of an alarming political deterioration and a progressive breakdown of trust in institutions.

At the heart of that rupture are hundreds of thousands of French people who experience the same thing every day: a tax administration that has become so complex, so procedural, so disconnected even from case law, that it crushes those who are simply trying to understand their rights.

My testimony

I am a private landlord in Auvergne. I have conscientiously paid my taxes for twenty years and have never been reassessed. Today, I am in a pre-litigation phase with the tax administration over three fiscal years, despite all the provisions of the General Tax Code I have produced, despite the Council of State rulings I have cited, and despite the applicable case law I have documented.

As in a significant share of cases that eventually reach an administrative court, I know statistically that the administration may ultimately be overruled. But in the meantime, the taxpayer endures months of procedure, tax-lawyer fees, considerable psychological pressure and sometimes a legal mortgage on his property.

The system has gone mad

The mere processing of a disputed tax file now mobilizes a heavy administrative and judicial chain: public-finance officers, mediation, administrative court, registry, rapporteur, judge, sometimes appeal. Often for a dispute that should have been resolved by a correct reading of the law and of case law.

Proceedings before an administrative court can last 18 to 36 months. During that time, the State continues to collect, surcharge and penalize, even when the taxpayer may be right. The psychological cost is immense: anxiety, sleepless nights, legal research, fees, and the feeling of being treated as presumed guilty by one’s own Republic.

The official figures of this organized waste

The French tax administration mobilizes around 105,000 agents if one takes the broad perimeter of the fiscal and financial sphere. Control, litigation, management, collection, reception and mass processing absorb a considerable amount of human energy.

The problem is not the existence of agents. The problem is the organization of a machine that mobilizes too many people on repetitive tasks, slow procedures and administrative readings sometimes contradictory with case law.

And yet the solution exists

Generative artificial intelligence and legal expert systems could process in a few minutes what now takes months:

• automatic analysis of the General Tax Code and Council of State case law;
• real-time, reasoned and sourced response to the taxpayer;
• automatic detection of cases in which the administration contradicts case law;
• prior administrative appeals processed in a few days instead of several months;
• a final decision within 30 days maximum, with a human magistrate for complex cases.

The potential economic impact would be massive: elimination or redeployment of a large share of repetitive tax-litigation tasks, budgetary savings, reduction of congestion in administrative courts, and refocusing of human agents on complex fraud and support for vulnerable taxpayers.

The model already exists abroad

Estonia, Sweden, the United Kingdom and Singapore have shown that a modern tax administration can be simpler, faster, more automated and often better accepted. The question is therefore not whether transformation is possible. It is why France keeps delaying the obvious.

Union inertia and the paradox of the agents

The agents are not the problem. Many of them suffer from the archaic system themselves: overload, loss of meaning, burnout, contradictory injunctions, outdated software, heavy procedures and lack of institutional discernment.

The paradox is total: taxpayers suffer, agents suffer, but intermediary structures too often defend the maintenance of headcount and procedures instead of demanding the liberation of agents through automation of repetitive tasks.

The real fight is therefore not against civil servants. It is against the machine that traps them as much as it crushes citizens.

When the administration refuses to apply the spirit of court decisions

The most serious scandal concerns the coherence of the rule of law. When case law exists, the taxpayer should not have to fight the same battle for months to have an identical situation recognized. A modern administration should automatically integrate decisions of the Council of State, apply them to comparable cases and avoid unnecessary litigation.

Artificial intelligence could once again instantly detect contradictions between administrative doctrine, applicable texts and recent case law.

The Benjamin Brière case

The Benjamin Brière case spectacularly illustrated this dysfunction. A former French hostage detained in Iran, he had to face an administrative mechanism incapable of properly handling an exceptional human situation. This was not a mere incident. It revealed a system that applies its protocol before understanding the person.

A properly designed AI, correctly connected to the relevant databases, could have detected the anomaly, cross-checked the data, triggered a priority regularization procedure and avoided the administrative humiliation.

This is the absolute paradox: AI, often accused of being dehumanizing, could have been more human than the current machine, because it would have been designed to recognize exceptions.

Why I am publishing this analysis

I am publishing this testimony consciously and in the general interest. My objective is not to settle a personal dispute, but to issue a citizen’s warning about a systemic dysfunction of the French State.

I reveal no confidential data. I name no individual agent. I describe a system, its figures, its dysfunctions, and I propose concrete solutions inspired by European models that work.

If France can no longer accept that an informed citizen publicly describes the dysfunctions of his administration, then France is no longer a democracy.

It is no longer our role to remain spectators of decline. It is our responsibility to change course democratically, before the markets or the streets do it in our place.