Asma Mhalla, political scientist, publishes on September 20, 2026 on her newsletter Dangerous Ideas an essay written on returning from a European tour, during which she moved within the same hour from artificial intelligence to the AfD and to sovereignty. Her hypothesis: September 2026 tells a single story whose pieces have not been connected.

On one side, the AI sequence. On September 9, Jacob Coxon, a young researcher who passed through OpenAI then Anthropic, resigns and publishes a viral warning: advanced AI could kill humanity before the decade is out. A week later, Geoffrey Hinton comes out of a briefing to Congress and estimates that there may be only a year left to put guardrails in place. Industry leaders diverge publicly: Amodei calls for slowing down, Altman demands more oversight, Huang refuses any coordinated slowdown. On the other side, the ballot box: the AfD obtains 43.8% in Saxony-Anhalt, Roberto Vannacci's Futuro Nazionale settles around 7.5% in Italy, Marine Le Pen gathers 34 to 36% of voting intentions in the first round.

Mhalla refuses to take a position on the probability of extinction and refers to the rapport international de sécurité de l'IA: current capabilities insufficient for a loss-of-control scenario, rapid progress, divided experts. Her real subject lies elsewhere: what fear manufactures. A word circulates everywhere — replacement — because before being a theory, it denotes a sensation: something could carry on without me. AI is reaching the professions that educated classes believed protected, and the 2026 Edelman Trust Barometer measures French disenchantment: 6% think the next generation will live better.

Then comes retrofuturism. Futuro Nazionale, MAGA's "again": the desirable future is a restoration. Drawing on Jeffrey Herf's Reactionary Modernism, she recalls that the authoritarian right never fought technology; it tore it away from the market to hand it to the nation. Technology is spoils, and the only question is who it belongs to. When its producers declare they no longer know how to govern it, they hand the forces of restoration their favorite vocabulary. Not an electoral causality: a narrative compatibility.

France illustrates the paradox through a volatile doctrine of sovereignty, of which Palantir at the DGSI, "transitional" since 2015, is the example. Conclusion: fear is cheap, it excuses one from the work. Serious AI risk calls for documented incidents, independent audits, reproducible evaluations, public investment. "I understand this fear. I refuse to grant it the status of thought."