The Lafon report: toward a state algorithm for administering digital truth

On July 8, 2026, the French Senate’s culture committee, granted for the occasion the investigative powers of a parliamentary inquiry, registered information report No. 875 on the regulation of information in the digital space. Three rapporteurs, Laurent Lafon, Agnès Evren, and Sylvie Robert; fifty-six recommendations; and the very first of them contains a proposal worth reading twice: create, before the next presidential election, a disinformation observatory tasked with prompting platforms to modify their algorithms or to “render invisible a user at fault” as elections approach.

At fault for what, judged by whom, with what right of appeal? The report never says. And that silence is precisely what structures all one hundred and fifty pages of the document. Nowhere does anyone propose banning speech. What it proposes is better: administratively organizing what will be visible, funded, recognized, and recommended. Everything else will still exist, somewhere, unfindable.

A largely accurate diagnosis, and that is exactly the problem

Let’s be fair to the text, because this is where its strength lies. The report’s economic assessment is solid, well documented, and hard to dispute. The French digital advertising market reached 12.4 billion euros in 2025, 76 percent of which was captured by eight international players who absorb 83 percent of the market’s annual growth. Press advertising revenue has been cut in half since 2012. The cost of producing political and general-interest news is estimated at 2.9 billion euros a year, 70 percent of it payroll. The value created by news is captured by intermediaries who produce none of it and bear none of its costs. All of that is true.

But what the report describes without naming it is textbook creative destruction. A rent-based model built on the scarcity of distribution channels is collapsing because the scarcity is gone. The legislator concludes that the market structure must be frozen in place through subsidies and regulatory privilege. Yet the report contains the refutation of its own logic: it cites Mediapart, with 270,000 subscribers and 150 employees, profitable, with no advertising, no dependence on shareholders, no government window to knock on. The proof that demanding journalism can live off its readers sits inside the very document that concludes it must be made to live off the state.

The rhetorical sleight of hand happens in a single sentence, always the same one: the business model is in peril, therefore liberal democracy is under threat, therefore anything becomes legitimate. That is the argumentative bridge on which all fifty-six recommendations rest. Once you spot it, you can’t unsee it.

The observatory that makes people invisible

Recommendations 1 through 4 form the operational core of the apparatus. The disinformation observatory would be the counterpart of Viginum, the state’s technical service responsible for monitoring foreign digital interference, but aimed at manipulation “of domestic origin.” Let’s pause on that phrase. Viginum was created to detect information operations run from abroad, and on that front the report praises its work during the 2026 municipal elections. What is now being proposed is to turn the tool inward. An apparatus designed against Russian intelligence services will be aimed at French citizens.

Then there is invisibilization itself. It is not content removal, a decision that is notified and can be challenged. It is punishment without trial, without notice, without a judge, and without appeal, carried out by a private third party at the prompting of a body fed by “civil society,” that is, by associations and researchers whose democratic legitimacy no one has ever validated. I once described punishment without judgment at the European level as digital lettres de cachet, the sealed orders by which the old French monarchy imprisoned men without trial. The Senate now proposes the national version: homespun, and outsourced to advocacy groups.

The timing completes the picture: “as elections approach,” which is exactly the period when a citizen’s voice carries the most weight and deserves the most protection. Recommendation 4 goes further still and contemplates suspending recommendation systems during the campaign. Take the measure of that admission: the state now regards the architecture of the digital public square as an adjustable electoral parameter. You don’t stuff the ballot box; you adjust the volume.

Defining truth by regulation

Recommendations 14 through 17 take aim at European law. The idea is to write an explicit definition of “disinformation” into the Digital Services Act, to distinguish it from “malinformation,” and then to grade the risks on the model of the AI Act: low, high, unacceptable, with earmarked categories for climate and health.

Malinformation first, because it is the most revealing concept in the entire corpus. It designates information that is accurate but deemed harmful because of its context or its presumed intent. Truth itself becomes punishable. No free society has ever built anything good on that idea.

Then the legal definition of disinformation. To define in law what is false is to define, in negative space, what is true, and to hand that definition to committees. Recent history is littered with reversed consensuses: hypotheses treated as conspiracy theories and banned from platforms before becoming admissible lines of inquiry debated in scientific journals. Yesterday’s “unacceptable risk” is tomorrow’s authorized opinion. A law that freezes the boundary also freezes the errors of the moment it was written.

As for the earmarked categories, climate and health, they designate without saying so the domains where debate will be most tightly administered, which is to say precisely those where public policy commits the most coercion and the most money. The gray zone the Senate claims to clean up, it merely relocates into its own definitions.

The offense without intent and the algorithmic publisher

The criminal-law section deserves a careful read, because it contains two ruptures.

Recommendation 5 proposes opening civil action for the false-news offense under Article 27 of the 1881 Press Law to any association duly registered for at least five years. Translation: the activist judicialization of public speech, with each camp bankrolling lawsuits against the other. The delicious part is that the same report devotes its Recommendation 55 to fighting SLAPP suits, those abusive legal actions designed to financially exhaust journalists and whistleblowers. In Recommendation 5 the Senate manufactures the very weapon it outlaws in Recommendation 55. Apparently, no one proofread the whole thing.

Recommendation 9 is the second rupture: creating unintentional variants of defamation, insult, and discrimination for cases where these harms result from an artificial intelligence system and mere negligence. Since 1881, French press law has rested on intent. This introduces liability for carelessness. For any publisher who uses AI in his production chain, and I am one, it installs a diffuse criminal risk whose only rational hedge is preemptive over-moderation. Hardly anyone will be prosecuted; everyone will police himself. It’s cheaper that way.

Recommendations 18 and 19 complete the edifice by seeking to make platforms and AI model providers editorially liable for the configuration of their algorithms. The idea is seductive: hold the giants accountable, and who could object? But a legally liable publisher moderates anything that carries the slightest litigation risk. Editorial liability imposed at platform scale is not a guarantee of quality; it is a conformity machine whose effects will fall first on marginal, heterodox, inconvenient voices. Mine, perhaps yours.

A state-issued press card

Recommendations 20 through 26 overhaul accreditation by the CPPAP, the joint commission for press publications and agencies whose stamp of approval governs subsidies, tax treatment, and official recognition. The proposed new criteria: a minimum proportion of professional journalists, adherence to ethics charters, possible consultation of an “accredited association active in matters of journalistic ethics” (Recommendation 22, which leaves wide open the question of who accredits the accreditor), and the absence of “disproportionate” use of generative artificial intelligence without “genuinely journalistic” treatment (Recommendation 26).

Disproportionate. Genuinely. Two adverbs, zero thresholds. This vagueness is not sloppy drafting; it is the instrument itself. An undefined criterion hands the licensing window discretionary power, and discretionary power breeds docility in those who depend on it. The report even has the elegance to restate the founding principle on its own: taking a stand is a constitutive feature of press freedom, protected by the law of 1881, and the French press was historically built on external pluralism, the coexistence of publications with sharply drawn editorial lines. It then devotes seven recommendations to rebuilding, through economic means, the regime of prior authorization that the law of 1881 had precisely abolished.

The anointed influencers and everyone else

If the Senate means to lock down access to state aid, it also comes straight for those who chose to do without it. So I read Recommendations 36 through 40 from a particular vantage point: that of the party concerned. This blog counts more than 480 pieces of analysis and opinion, with no CPPAP accreditation and no desire for one. I also run an automotive YouTube channel. I stand on both sides of the line this report sets out to draw, and never on the subsidized side.

Because what these recommendations organize is, quite plainly, a sorting. On one side, the audited and applauded news creators, HugoDécrypte, Gaspard G, Jean Massiet, cited by name, whom the report proposes to include among the “services of general interest” that platforms would be required to promote algorithmically, complete with dedicated public funding and a pay scale for their journalists. On the other, the mass of ordinary creators, the most audible of whom, under Recommendation 38, would fall under the direct regulation of Arcom, France’s audiovisual and digital regulator, whenever their program is “similar” to a television program and presents “serious risks.” Similar, serious: the same vagueness technique, this time applied to broadcasting. I have already written what I think of this authority’s drift and of its appetite for regulating what exceeds it; this report hands it exactly the expansion of territory it was waiting for.

The result is a three-tier hierarchy of speech: at the top, the creator of general interest, promoted by algorithm on public instruction and financed with public money; at the bottom, the common-law creator, exposed to the hardened criminal regime described above; and in between, a licensing window that decides who moves up. And then there is the irony the report records without drawing the lesson: the three creators it heard unanimously rejected any form of labeling, in the name of free expression and the risk of raising barriers to entry. The system’s first intended beneficiaries want no part of it. It will be built anyway.

Funding the press with its regulators’ fines

Recommendation 46 proposes creating a special budget account fed by the fines levied by CNIL, the French data protection authority, and by Arcom, for the benefit of the press. Let’s spell out the circuit slowly, because its beauty must be earned: the authority imposes sanctions, the proceeds of the sanctions fund the media, and the media cover the authority’s work. A structural conflict of interest, not tolerated as an imperfection but enshrined as a permanent budgetary mechanism. The harder the regulator strikes, the better the press does, and vice versa.

Recommendations 28, 29, and 53 round out the scheme on the advertising side. First, transparency about the sites where ads are served, and the exclusion of “disinformation sites” from government ad buys. Then, for private advertisers, an escalation the text openly embraces: rewarding investment in news media, providing an incentive mechanism, “or even an obligation if transparency is not enough to make advertisers accountable.” Incentive, transparency, obligation. In three words, the freedom to allocate private advertising capital becomes an instrument of information policy. Steering advertising flows toward approved media by force is nationalizing the press-financing function without ever owning the word or paying its political price.

From outsourcing to Brussels to bringing it home

This report has to be placed on a trajectory. I wrote in February that France had outsourced its censorship to Brussels, the DSA offering national governments the comfort of repression without the responsibility. I added in April that the Union now punishes without judging. Report No. 875 closes the loop: France equips itself with its own domestic machinery, observatory, licensing windows, accreditations, special accounts, all while asking Brussels to tighten the DSA further, to define disinformation within it, and to extend its obligations to AI platforms. Outsourcing becomes vertical integration.

And the text deserves credit for one moment of involuntary lucidity: its danger is not brute censorship, and it very nearly says so itself. No one will be forbidden to speak. Some will simply be unfindable, unfunded, unrecognized, unrecommended, while others are promoted by algorithms configured on public instruction. A political administration of visibility needs no scissors. It needs licensing windows, vague criteria, and time. This report supplies all three.

Pluralism cannot be planned

The document quotes one of its own witnesses acknowledging that pluralism has probably never been in better health in France than it is today, at least on social media. It concludes that pluralism must therefore be regulated. A Senate that laments citizens’ distrust of the news proposes to answer it by labeling speech, subsidizing the compliant voices, and rendering the others invisible. It will not restore trust: it is methodically manufacturing the next distrust, the one that will strike the anointed media precisely because they were anointed.

The alternative exists, and the report supplies the evidence for it without seeing it. Media that live off their readers rather than off a licensing window. Creators who refuse the labels others want to pin on them. Citizens treated as adults capable of sorting for themselves, rather than as flows of attention to be steered toward content of general interest. The quality of information has never been decreed by a joint commission. It is earned, reader by reader, and it is lost the same way. Everything else is planning, and the planning of truth goes by a name the rapporteurs know very well, since it took them fifty-six recommendations to avoid writing it down.