French voters want AI rules. Their leaders are hitting the brakes

French voters want AI rules. Their leaders are hitting the brakes

8%. That is the share of French respondents who want to speed up AI development, according to an OpinionWay poll released by the Center for AI Safety (CeSIA) in April 2026. It is also, give or take, the one number in the survey that lines up with what their leaders just decided.

Because while 8% of respondents ask to floor the accelerator, the European Union has just pushed back the AI Act's heaviest obligations by about two years. A delay France actively lobbied for. The gap here is not only about the calendar. It is democratic. And it runs less between Brussels and Paris than between a majority of citizens and their own political class.

What the French want

The CeSIA poll covers a nationally representative sample of 2,065 adults, surveyed from March 25 to April 9, 2026. The numbers leave little room for interpretation.

42% of respondents want to slow AI development sharply or pause it. 56% would rather the state set the direction, against 25% who favor private companies. And 62% want the technology steered toward safety, rights and ethics, versus 21% for performance and innovation.

The most striking figure: 78% support international agreements to ban AI uses that threaten human life or fundamental rights. That majority cuts across the entire political spectrum. From 74% among far-right RN voters to 89% among conservative LR voters, taking in 79% at left-wing LFI, 79% among the Greens, and 88% at the Socialists as well as the presidential bloc. On few topics do you find agreement stretching that far end to end.

What their leaders decided

The AI Act technically still applies from August 2, 2026. But its core, the obligations for so-called high-risk systems, slides by roughly two years, to December 2027, and up to August 2028 for AI embedded in medical devices. The very definition of "high risk" was tightened along the way. We laid out the exact mechanics, the exemptions and the lobbying around them, in a separate piece.

The point sits elsewhere: this delay was carried by the capitals themselves, France first among them. The decision-makers hitting the brakes also sit in Paris, among the officials voters sent there.

Who is pushing, and in whose name

So who is pushing? As early as November 2025, France and Germany, two of Europe's biggest AI investors, backed the simplification, tied to the Commission's "competitiveness compass." In July 2025, a letter signed by 56 European companies, including France's Mistral and Germany's Aleph Alpha, was already calling to "stop the clock."

The argument is sovereignty and competitiveness: don't hamstring the continent's champions against American and Chinese giants. It carries weight. It also illustrates an openly porous line between public decision and industry, embodied most visibly by Cédric O, the former digital minister turned shareholder and adviser to Mistral.

The sovereignty that regulates and deregulates

There is an irony of timing. In late June, the EU was celebrating the naming of Europa, its future sovereign large model, in the name of the continent's technological independence (a project we looked at closely). A few weeks earlier, that same sovereignty was being used to justify delaying the rules.

The word works both ways. Yesterday, you had to regulate so as not to depend on foreign platforms. Today, you have to deregulate to let your own breathe. Sovereignty has become a variable-geometry argument, deployed to fit whatever outcome you are after.

What remains is the gap. On one side, a cross-partisan public asking for more guardrails. On the other, the governments it elected, delaying their rollout under pressure from their own industry. You can find the simplification legitimate or reckless, the debate is real. But the irony fits on a single line: on a topic where 78% of French voters agree, from the far right to the Socialists, it is their own elected officials stepping on the brake.