The national observation is damning, and it is officially documented:
The confidence of the French in the State for "the good use of public funds" fell from 33% to 22% between 2023 and 2025 (Barometer of the Council of Compulsory Levies, Harris Interactive, November 2025). That is a fall of 11 points in two years, unprecedented.
78% of the French judge the level of taxation too high, 75% now consider that it can be justified to expatriate in order to reduce one's levies (against 28% who judged this unjustified in 2023). Primary source: Barometer of tax and social levies 2025 (Court of Auditors / Council of Compulsory Levies / Harris Interactive), published on 27 November 2025.
59% of the French declare themselves ready to see the level of public services fall in order to lighten their tax pressure (IFRAP, February 2026).
27% of young people aged 18 to 25 consider that it is justified to cheat on taxes if one has the possibility (annual report of the Court of Auditors 2025, Senate hearing). An entire youth dropping out of the republican fiscal contract.
44% of the tax-fraud cases transmitted to the prosecution in 2024 were dropped without further action, only 27% were prosecuted before the court (Court of Auditors, report "The fight against tax fraud", December 2025). The repressive system itself is broken down.
The Court of Auditors itself qualifies in December 2025 French tax fraud as a phenomenon "poorly understood, poorly quantified, poorly treated" (public report, official presentation to the Senate). The machine no longer even knows how to measure what it is supposed to combat.
What we are collectively living through is not a mere "fiscal exasperation". It is a progressive rupture of consent to taxation, officially diagnosed by the Council of Compulsory Levies itself, which speaks of an "alarming deterioration on the political level" and of a "progressive rupture of confidence in the institutions".
And at the heart of this rupture, there are hundreds of thousands of French people who live daily the same experience: that of a tax administration become so complex, so procedural, so disconnected from case law itself, that it crushes those who simply try to understand their rights.
I am going to testify to this personally today, because my case is not isolated. It is the incarnation of a systemic phenomenon that threatens the very unity of the French republican pact.
My Testimony
I am a landlord in Auvergne, I have paid my taxes conscientiously for twenty years, I have never had a reassessment. Today, I am in a pre-litigation phase against the tax administration, over three financial years (2022-2024), despite all the texts of the General Tax Code that I have produced, despite the rulings of the Council of State that I have cited, despite the applicable case law that I have documented. "We apply the tax code and only the tax code". Translation: they do not take into account court decisions, whatever they may be, that concern other identical cases.
As in 35 to 40% of the cases that go up to the Administrative Court, I know statistically that the administration will end up being disavowed. But in the meantime, I will have undergone 18 to 36 months of procedure, tax-lawyer fees, and potentially a legal mortgage on my property.
The system has become mad:
The mere processing of a file like mine now mobilises a principal inspector of public finances, two controllers, the mediator of Bercy, an administrative magistrate, a clerk, a judge-rapporteur, and soon a government commissioner. Cost for the taxpayer: several tens of thousands of euros of public payroll mobilised, for a dispute that should have been settled in two letters.
The procedure before an administrative court today lasts 18 to 36 months (Court of Auditors, 2024 report). During this time, the State continues to levy, to increase, to penalise, while the taxpayer is in the right.
The psychological cost is massive: months of anguish, sleepless nights, hours of legal research, lawyer's fees when one has the means, the feeling of being treated as a presumed guilty party by one's own Republic.
The official figures of this organised waste:
The French tax administration mobilises about 105,000 agents (DGFiP, 2024 activity report), of which about 35,000 on control and litigation functions.
Total annual cost of tax control and tax litigation: estimated between 3.5 and 4.5 billion euros per year (Court of Auditors 2023).
Rate of rejection of prior administrative appeals in favour of the taxpayer: only 17% (DGFiP). But when the case goes up to the Administrative Court, the annulment rate rises to 35-40%.
French administrative courts: a backlog of more than 220,000 files in progress (Council of State, 2024), of which about 40% of a fiscal nature.
And yet, the solution exists and it is mature.
Generative AI could process in a few minutes what today takes months: automatic analysis of the texts of the General Tax Code and of the case law of the Council of State; reasoned and sourced response in real time to the taxpayer; automatic detection of the cases where the administration is in contradiction with case law; prior administrative appeal processed in a few days instead of several months; final decision in 30 days maximum, with a human magistrate on the complex cases.
The potential economic impact: elimination of 60 to 70% of the tax-litigation workforce (about 20,000 to 25,000 posts), redeployed towards the major international frauds; savings of 2 to 3 billion euros per year; reduction of litigation by 70 to 80%.
The model already exists abroad: Estonia, tax declaration in 3 minutes, AI litigation in 5 days, economic impact quantified at 2% of GDP per year; Singapore, taxpayer satisfaction of 96%; United Kingdom, 85% of disputes settled online in 7 days; Sweden, minus 30% of workforce in 14 years, without social drama; Germany, OZG-Gesetz plan of complete administrative digitalisation by 2027.
The Documented Union Inertia And The Paradox Of Agents Who Suffer As Much As The Taxpayers
The DGFiP has already lost 50,000 jobs between 2002 and 2024 (source: CGT Public Finances). But this reduction was made at constant structure, without systemic modernisation, without an overhaul of the missions, without AI. The remaining agents each process more files, in a system still just as archaic, and therefore suffer more.
The main union organisations of the DGFiP have organised on average 4 to 6 national strike days per year since 2020. None of these calls to mobilisation bears on the modernisation of the missions or the introduction of AI. All bear on remuneration, working conditions, the maintenance of the workforce.
The NRP (New Proximity Network), the main reorganisation project launched in 2019, is systematically fought without any modern structural alternative being proposed by the unions.
The "Without Us" campaign (Solidaires Public Finances, 2024) affirms that "without the physical agents, the public service collapses". It is exactly the inverse argument to the Estonian, Swedish or British modernisation, which demonstrates that digitalisation and quality of service go together.
The paradox is total: the agents suffer as much as the taxpayers. Burn-out, exhaustion, demotivation, loss of meaning are massively documented. But the intermediary bodies obstinately refuse the only reform that could free the agents: digitalise the repetitive tasks, redeploy the humans towards the high-value-added missions.
Result: everyone loses. Only the machine itself turns, in the void.
The Absolute Scandal: When The Tax Administration Refuses To Apply Court Decisions
Here is a little-known but massive mechanism, which on its own would justify an emergency reform.
When the Council of State renders a ruling in favour of a taxpayer on a question of interpretation of the General Tax Code, this ruling becomes a jurisprudential precedent that should impose itself on the whole tax administration. Yet the tax administration regularly refuses to apply these rulings to identical situations, opposing to the taxpayers a standard argument: "We are not the justice system. If you contest, go to the administrative court. In the meantime, we apply our interpretation and we take a legal mortgage on your property".
Consequences for the taxpayer: they must initiate a procedure before the Administrative Court simply to have confirmed what the Council of State had already settled in identical cases; they undergo a legal mortgage of the Treasury on their real estate for 18 to 36 months; they assume alone the tax-lawyer fees.
Consequences for the State: artificial overload of the administrative courts by litigation that the case law of the Council of State had already settled; budgetary cost estimated between 800 million and 1.2 billion euros per year; congestion of the administrative courts of appeal; a 3-to-5-fold multiplication of the total cost of a dispute that could have been settled upstream.
AI would resolve this in a few seconds: automatic detection of the cases where the administrative interpretation contradicts a recent ruling, automatic alert to the inspector, proposal of a decision conforming to case law, saving of 18 to 36 months of useless procedure.
The real subject is not only administrative productivity. It is the coherence of the rule of law itself.
The Benjamin Brière Affair: When Media Coverage Became The Only Protection Of Citizens Against The Machine
On 9 April 2026, the country learns a story that should have been the stuff of Kafkaesque fiction.
Benjamin Brière, a former French hostage detained for 1,079 days in Iranian prisons (2020-2023), was summoned on his return to France by the tax administration, which reproached him for not having made his tax declarations during his years of captivity. The agent of the tax service reportedly replied to him, remarks reported by France Info, Le Journal du Net and officially acknowledged by the DGFiP: "Even in prison, one makes one's declaration."
Then, when Mr. Brière specified that he was in an Iranian prison and that he had no right to communicate with his family: "In that case, your family could have done it."
Beyond the tax office, Mr. Brière was also "struck off everywhere": Social Security, France Travail, elementary administrative rights. He had to fight alone, for months, to reactivate his French administrative existence.
What finally unblocked the situation was neither case law, nor good will, nor an institutional protocol. It was media coverage, and the personal intervention of a deputy. On 12 April 2026, the Directorate General of Public Finances published official apologies on X, acknowledging "an abnormal first contact" and an "unacceptable" response. Mr. Brière's situation was regularised without any penalty, for the four years of non-declaration.
But here is the real question that this affair poses: how many anonymous French people, without a deputy to defend them, without a media platform, without a lawyer at 500 euros an hour, undergo daily the same type of mechanical response? How many citizens "struck off everywhere" by an administration that does not provide a box in its form for their exceptional situation? How many cases like this one are never covered by the media, and are resolved by exhaustion, indebtedness, or even fiscal exile?
The Brière affair is not an isolated dysfunction. It is the most media-covered incarnation of a systemic mode of functioning: the administration applies its protocol, mechanically, without discernment, because its agents are trained for that, because the databases are not interconnected, because no box "hostage", "long-term sick", "victim of natural disaster", "expatriate for medical reasons" exists in the forms.
And that is exactly where well-designed AI would have made the difference: a correctly trained AI would have automatically cross-checked the databases of the Quai d'Orsay, of Social Security, of France Travail and of the DGFiP; it would have detected the anomaly from the first consultation of the file; it would have triggered an automatic procedure of regularisation without penalty; it would have generated an official welcome letter.
It is the absolute paradox: AI, accused by many of being dehumanising, would in reality have been more human than the current machine, because it would have been correctly designed to recognise the exception, cross-check the data, and treat with dignity the out-of-the-ordinary citizens.
Today, in France, one must be covered by the media to be correctly treated by one's administration. It is the sign of a democracy that no longer functions.
To The Agents Who Read Me
To the agents who read me and who recognise themselves in this observation: you are not the problem. You are the first victims of this machine become mad, incapable of adapting to a world that evolves at high speed.
The problem is not to know whether this or that agent is individually guilty. The problem is that thousands of posts today produce, structurally, a mediocre result: slowness, mechanical responses, ignorance of case law, useless litigation, disproportionate public cost. When a system massively produces bad results, it must not be protected: it must be eliminated or recast.
One does not reform a machine that malfunctions by protecting all its cogs. One begins by identifying those that no longer serve, those that cost more than they bring in, those that technology can replace with more speed, coherence and legal security. Then one eliminates them.
Let us be clear: AI must not serve to repaint in digital an administration become inefficient. It must allow the elimination of the administrative posts become objectively useless when their mission consists only in mechanically applying forms, ignoring case law, moving files around, producing standardised responses or making citizens wait for months.
It is not a matter of artificially conserving thousands of posts in the name of a misplaced compassion. When a task can be accomplished faster, more justly, more clearly and at lower cost by a correctly controlled AI, then the corresponding post must disappear. It is brutal to hear, but it is the very condition of the recovery of the State.
One must cease to repeat that the human must be conserved as soon as a file becomes complex. It is often the inverse: the more a file is complex, the more AI becomes useful. The major international frauds, the patrimonial investigations, the cascading legal arrangements, the financial flows between companies, the fiscal incoherences, the cross-border situations and the repetitive litigation are precisely the domains where a well-designed artificial intelligence can analyse faster, cross-check more widely, spot more finely and reason more rigorously than a saturated human service.
The human role must therefore no longer be to do by hand what the machine can do better. It must be limited to what requires a political responsibility, a moral appreciation, a direct contact with fragile citizens, a human mediation or a final validation of the most sensitive decisions. The accompaniment of the elderly, the handicapped, the isolated, the digitally illiterate or the psychologically overwhelmed by the administration must remain human. But the fiscal analysis, the comparison with case law, the detection of frauds, the patrimonial investigation, the preparation of decisions, the reasoned drafting of responses and the sorting of litigation can, in the immense majority of cases, be entrusted to AI.
This means one very clear thing: a large part of the current posts no longer has a vocation to be maintained. Not out of hatred of the agents, but because the mission itself becomes technologically obsolete. When an AI can produce in a few minutes an analysis more complete, more sourced, more coherent and more conforming to the law than an administrative service in several months, the maintenance of the post is no longer a social policy: it is a public waste.
The argument consisting in saying that the agents should be redeployed towards the complex files is largely outdated. The complex files are precisely those where AI can produce the most value: mapping of patrimonies, detection of shell companies, analysis of international flows, automatic comparison with case law, spotting of administrative contradictions, hierarchisation of the fiscal risk, reasoned drafting of decisions. What the human did slowly, partially and sometimes contradictorily, AI can do massively, quickly and with a superior traceability.
The maintenance of human agents must be reserved for very limited functions: reception of vulnerable persons, mediation with fragile taxpayers, ethical control, final validation of serious decisions, political and jurisdictional responsibility. For the rest, the State must assume the progressive elimination of the administrative posts become useless. France will not recover by indefinitely protecting functions that technology renders obsolete.
France can no longer indefinitely finance thousands of public jobs whose real function consists in slowing down the country, exhausting the taxpayers and producing useless litigation. AI must therefore be assumed for what it is: not a modernisation gadget, but a tool of massive reduction of the administrative workforce become obsolete.
An Assumed Whistleblower Posture
I publish this testimony in full awareness and in the general interest, in application of the principles of the Sapin 2 law (2016) and of the law of 21 March 2022 improving the protection of whistleblowers. My objective is not the settlement of a personal dispute, which follows its normal procedural course, but the citizen alert on a systemic dysfunction of the French State, documented by the official figures of the Court of Auditors, of the Council of State, of INSEE and of the DGFiP itself.
I reveal no confidential data. I name no agent individually. I describe a system, its figures, its dysfunctions, and I propose concrete solutions inspired by the European models that work.
If France can no longer accept that an informed citizen publicly describes the dysfunctions of its administration, then France is no longer a democracy. I trust the French rule of law to respect this citizen speech.
Why I Publish This Analysis, And Why I Am Going To Continue
I am neither a journalist, nor an elected official, nor a candidate, nor a party militant. I am simply a French citizen of 50 years, a landlord in Auvergne, a self-builder of a bioclimatic house, a YouTube content creator on autonomy and sovereignty, who made a lucid observation a few years ago: France is sinking, and we no longer have the luxury of waiting for the political leaders to notice it in our place.
So I made a decision: to mobilise my personal time and my own financial resources to produce a series of documented citizen dossiers on the reform of the French State. Institutional sources only: Court of Auditors, Senate, Council of State, INSEE, OECD, Eurostat, Légifrance. No opinion, no ideology, just official figures, rigorous international comparisons, and concrete costed proposals.
These dossiers exist, and some are already freely accessible:
Dossier I, Category A "Agencies to keep": 552 pages, 552 entities identified, 30 to 56 billion euros of documented annual savings, completed on 8 May 2026.
Dossier II, "The merger of agencies": 1,020 pages, 35 clusters of merger documented, 2.3 to 4.1 billion euros of annual savings, completed on 12 May 2026.
Dossier III, Category C "Agencies to reintegrate": in the process of finalisation.
Dossier IV, Category D "Agencies to eliminate": in production.
Dossier V, "Implementation kit": in production.
Global objective documented line by line: 350 billion euros of potential annual savings, over a perimeter of 1,494 agencies and operators of the State identified by the official sources: senatorial report 807 of 3 July 2025 "Barros-Lavarde", budgetary Yellow of operators PLF 2025, list of the Budget Directorate 24 February 2025.
Why do I do this? Because 2027 is a capital deadline for France, and because the public debate deserves better than slogans and uncosted promises. Because the debt reaches 117.7% of GDP, the interest on the debt becomes the first budget of the State at 74 billion euros in 2026, and no one can seriously claim that this trajectory is sustainable. Because the country is on the edge of social revolt, and because the responsibility of informed citizens is not to keep quiet but to propose constructive paths before the street or the markets decide in our place.
What I propose is not a candidacy. What I propose is a citizen contribution to the national debate, in open access, so that each French person, each public agent, each elected official, each journalist, each unionist, each entrepreneur can seize it, criticise it, improve it, prolong it.
If even a single one of these dossiers inspires a concrete reform by 2030, my citizen commitment will have been useful. If none bears fruit, I will at least have done my part. But I can no longer remain a spectator.
The whole of the dossiers and articles is freely accessible on delta-sierra.com.
It is no longer for us to be spectators of the decline. It is for us to right the ship democratically, before the markets or the street take care of it in our place.
Mother Dossier of 300 pages for nearly 350 billion euros of annual savings, available on Amazon: https://www.amazon.fr/dp/B0H3FN54S2
Find my articles and my books on my website: https://www.delta-sierra.com/index.html
David SALVAN, sole trader, Coutansouze, Allier. Self-builder | Private landlord | Productively angry citizen.