France is reshaping its AI strategy in 2026—pushing for “digital sovereignty,” watching the assistant-AI market fragment, and preparing for new European bans aimed at abusive image tools.
After years of relying on Palantir’s analytics tools inside France’s domestic intelligence service, the DGSI (Direction générale de la sécurité intérieure), the government says it plans to switch to a French alternative. Sébastien Lecornu announced Palantir will be replaced by ChapsVision, a French company founded by Olivier Dellenbach. The transition will likely happen sometime in 2027 once the “technicalities” are resolved, according to the economy minister, who did not specify when Palantir would be fully shut off.
The shift is part of a broader political push highlighted at VivaTech, where European leaders and officials urged cutting dependence on U.S. technology after the Trump administration decided to restrict access to Anthropic’s models. ChapsVision is positioning itself as a European “trusted data and AI” champion at a moment when U.S. firms have long dominated the field.
ChapsVision vs. Palantir: a slow pivot in the name of “sovereignty”
Dellenbach’s company is not a newcomer to sensitive government work. The firm was already working with France’s Interior Ministry before the announcement, and it is selling itself on “trusted AI” designed for agencies that say they cannot risk placing sensitive data on foreign servers subject to U.S. law.
But the changeover is not imminent. As Le Figaro reported, a 2027 timeline leaves Palantir in place for months, and there has been no public, definitive comparison of the two systems’ technical performance. Palantir’s tools have had time to become embedded in day-to-day workflows—making replacement inside domestic intelligence more than a political decision.
The debate also reflects a broader tension Le Figaro identified: the biggest tech players are American and increasingly Chinese, while France points to world-class researchers supported by a strong math training pipeline. The ecosystem exists; the question is whether the state will place enough major orders to help French and European players compete with U.S. hyperscalers and American surveillance platforms.
ChatGPT dips below 50% market share as assistant AI becomes a three-way race
For three years, OpenAI dominated the assistant-AI market. That near-monopoly is fading. ChatGPT’s market share has fallen below 50% for the first time, according to Le Figaro, pressured by Google and Anthropic.
The shift changes how the sector is read: OpenAI is no longer the only company setting usage norms. Google’s Gemini and Anthropic’s Claude are gaining ground in a market that is becoming more fragmented as competition intensifies. For Mistral AI, the French player trying to carve out space, the leader’s erosion mechanically creates room—even if the resource gap with U.S. giants remains enormous.
This reshuffling has become directly political. Days after the Mythos 5 model was banned outside the United States, leaders from OpenAI, Anthropic, Google DeepMind, and Mistral AI attended a working lunch with heads of state from the G7 and partner countries—another sign that AI governance is now being negotiated at the highest diplomatic level.
Pourquoi l'IA française peine à s'imposer face aux Américains
EU sets Dec. 2, 2026 deadline to technically block “undressing” AI tools
The European Union has set a clear cutoff date: December 2, 2026. From that point, AI systems capable of generating images that digitally “undress” people must include safety measures that prevent that content. The rule applies to all AI systems sold on the European market.
The targeted ban is part of the EU’s phased rollout of the AI Act. It focuses on tools that have spread with little oversight and are often used to create non-consensual images of women from publicly available photos. The technical requirement is explicit: the system must refuse to generate the content, not merely display a warning.
For U.S. and Chinese platforms distributing these tools in Europe, it means forced compliance. For European developers building image models, it becomes a design standard to integrate from the start—an added cost, but also a potential selling point in markets where personal-data protection matters.
Gabriel Attal brings AI into France’s 2027 presidential campaign with a €200 billion plan
France’s 2027 presidential race is already shaping the tech debate. Gabriel Attal, the candidate of Renaissance, has pitched a “France 2040 plan for AI and innovation” priced at €200 billion (about $216 billion). He described it as “the equivalent of the Common Agricultural Policy” applied to artificial intelligence and listed AI among his four “major projects.”
Meanwhile, VivaTech hosted Jeff Bezos. The Amazon founder argued for an optimistic view of the digital revolution and said humanity will be “limited not by its capabilities, but by its imagination.” He also reiterated his ambition to establish a lasting presence on the Moon, despite the recent explosion of the New Glenn rocket.
The conference also spotlighted the 2026 French Tech Next40, whose selection signals a turn toward “breakthrough” technologies. AMI Labs, Aura Aero, Pasqal, and Wandercraft were among the 40 most prominent startups chosen by Bercy—France’s finance ministry—within an expanded list of 120 companies. The French ecosystem is real and increasingly structured, but it is evolving in a world where the rules are being negotiated between Washington, Brussels, and Beijing.
IA en France 2026 : souveraineté, marché et réglementation
- ChatGPT est passé pour la première fois sous les 50 % de parts de marché des assistants IA.
- ChapsVision, société française, remplacera Palantir auprès de la DGSI, probablement en 2027.
- L'Union européenne interdit les IA de déshabillage à partir du 2 décembre 2026.
- Gabriel Attal propose un plan IA de 200 milliards d'euros dans sa campagne présidentielle.
- Le French Tech Next40 2026 sélectionne AMI Labs, Aura Aero, Pasqal et Wandercraft parmi ses pépites.
Sources
3 sources · 6 faits sourcés
