A response to a documented threat, not a hypothetical one
On July 8, 2026, Prime Minister Sébastien Lecornu told the Senate he would bring a bill to the Council of Ministers by the end of July to triple the criminal penalties for producing false content during electoral periods, extend an emergency judicial takedown procedure from national elections to every local election, and create a permanent "public information commission" to flag interference. He was answering a question from Senator Claude Malhuret about AI-generated disinformation ahead of the 2027 presidential race — and he had a concrete case to point to, not just a worry about the future (Public Sénat).
Viginum, the state service that tracks foreign digital interference, confirmed four distinct foreign operations targeting the March 15 and 22, 2026 municipal elections, including a campaign dubbed "Rokh Solis" and pro-Russian networks that had previously targeted France (SGDSN/Viginum). One operation used AI-generated images and automated fake accounts to circulate false rape accusations against an LFI deputy via QR-coded flyers on campaign billboards in Marseille. It was the first election where the government's new cross-agency election-protection network, the RCPE, was operational — an acknowledgment that existing tools weren't built for this.
The case for acting now
The strongest argument for Lecornu's bill is that French law genuinely has a gap. The offense of spreading false news that skews an election currently carries just one year in prison and a €15,000 fine, and — crucially — it was written before generative AI existed, so it's unclear it cleanly covers deepfakes and synthetic voice clones. The emergency judicial referral that lets a court order fake content taken down mid-campaign, created by the 2018 loi relative à la lutte contre la manipulation de l'information, only applies in the three months before presidential, legislative, or senatorial votes — leaving France's roughly 35,000 municipal races and other local elections without the same fast-response tool, even though the March interference campaigns targeted exactly those races. Given that presidential-level threats were detected during a municipal election, tightening the law before the 2027 presidential campaign rather than after an incident is a defensible sequencing choice.
Why the mechanism, not the goal, is where scrutiny belongs
The 2018 law survived constitutional review only because the Conseil constitutionnel attached a specific, narrow limit to it: content can be ordered offline only when its falsity is "manifestly apparent" and its electoral impact is "apparent," and the law cannot reach opinions, parody, or exaggeration — only objectively verifiable false statements of fact (Conseil constitutionnel, décision n° 2018-773 DC). That reservation is the entire reason the referral procedure has coexisted with free expression for eight years without becoming a general censorship tool. Lecornu's extension multiplies the number of contexts in which that judgment has to be made correctly, fast, and under political pressure: thousands of hyperlocal races, often decided by margins where a rumor about a mayor's conduct is far harder to adjudicate as "manifestly false" within days than a claim about a presidential candidate that national media can quickly check.
The "public information commission," to be created by decree, is billed by Lecornu as an alert body for journalists, judges and citizens rather than a censor (TheNextWeb) — and a body that flags claims rather than orders removals is meaningfully less risky than one with takedown power. But "permanent" is doing real work in that sentence. The 2018 framework's legitimacy rests partly on its time-boxed nature: obligations activate in a defined pre-election window and lapse afterward, which Arcom's own guidance to platforms treats as central to the scheme (Arcom). A standing commission with no electoral-calendar trigger is a structurally different institution, and creating it by decree rather than through the same statute Parliament will debate for the rest of the bill means its mandate, funding, and removal from office won't get the same scrutiny as the penalty provisions.
What should happen in the autumn debate
None of this argues against tripling penalties for genuinely AI-fabricated electoral fraud, which is a proportionate response to a documented harm. It argues for Parliament to do three things the "very short" text as described does not yet guarantee: write the "manifestly false" and "apparent impact" standards from the 2018 law explicitly into the local-election extension rather than assuming courts will import them; give the permanent commission a statutory (not decree-only) charter with sunset review; and require Viginum-style public reporting on how often referrals are actually granted versus requested, so the law's real-world proportionality can be checked rather than asserted. Deterring foreign interference and protecting local political speech from state overreach are not in tension if the safeguards that made the 2018 law durable are carried forward — the risk is passing a broadly popular headline measure quickly and leaving that architecture to be improvised later.