France's spy agency is leaving Palantir — but the break isn't clean
France's domestic intelligence agency is replacing U.S. data giant Palantir with French startup ChapsVision.
But the American company’s contract still runs until 2028, and the French replacement won’t be deployed before 2027. Between political declaration and operational reality, the DGSI-Palantir case reveals the structural limits of digital sovereignty in Europe.
At a Glance
France’s domestic intelligence agency has relied on Palantir’s Gotham platform to process data since 2016 — its replacement by ChapsVision is not expected before 2027 at the earliest
The U.S. CLOUD Act of 2018 gives American authorities the legal power to demand access to data held by U.S. companies, even when stored abroad — an exposure France had no way to control under the Palantir arrangement
France and Germany are simultaneously replacing Palantir with the same platform, ArgonOS — a shared architecture that reframes the interoperability question entirely
This image is used for illustrative purposes only.
The announcement that contradicts a contract signed six months earlier
On June 16, 2026, on the sidelines of the VivaTech technology summit in Paris, French Prime Minister Sébastien Lecornu announced that the Direction Générale de la Sécurité Intérieure (DGSI) — France’s domestic intelligence and counterterrorism agency, roughly equivalent to the FBI’s national security division — would end its relationship with American data analytics giant Palantir and hand the contract to French company ChapsVision. “We cannot accept new strategic dependencies in the digital space,” Lecornu said, expressing his desire to build “genuine autonomy” rather than remaining subject to the goodwill of partners capable of “turning off the tap” on AI access.
The announcement was framed as a turning point. Politically, it is. But it runs directly into a contractual reality: in December 2025 — just six months before this declaration — the DGSI had renewed its contract with Palantir for three years, extending the relationship through 2028. Palantir responded to Lecornu’s announcement by stating in a press release that its contract with the DGSI remained “fully in force.”
This gap between announcement and contractual reality is not a minor detail. It reflects a tension that has run through French digital sovereignty policy for over a decade: the distance between stated political intent and the operational constraints that delay — sometimes indefinitely — the transitions that get announced.
What Gotham actually does — and why it posed a strategic problem
Palantir’s Gotham platform is not a file-management system. It is designed to cross-reference and analyze, in near real-time, massive volumes of heterogeneous data — phone intercepts, travel records, financial transactions, social media activity, administrative databases — to surface connections that human analysis alone would miss.
The DGSI began using Gotham in 2016, in the months following the November 2015 Paris terrorist attacks that killed 130 people. The decision was made under emergency conditions: at the time, according to the DGSI’s own public statements, Gotham was “the only tool available on the market capable of meeting the agency’s needs to address national security challenges.” The contract was renewed in 2019, again in 2022, and a third time in December 2025.
The problem was never the tool itself. It was what using that tool implied in terms of legal and strategic dependence. The parliamentary report n°2617, submitted on April 1, 2026, by National Assembly members François Cormier-Bouligeon and Aurélien Saintoul to the Defense Committee, was explicit: France does not control Gotham‘s source code, its underlying architecture, or its long-term maintenance conditions. In practical terms, if Palantir were to modify its access terms, unilaterally cancel the contract, or comply with a U.S. government order, France would have no autonomous capacity to respond.
That exposure is compounded by the CLOUD Act, a U.S. law enacted in 2018 that authorizes American prosecutors to compel U.S. companies to hand over data — including data stored outside American territory. Palantir is a California-based company. It falls squarely within the law’s scope. This means that data processed by the DGSI — among the most sensitive information held by the French state — could theoretically be subject to a U.S. legal demand. Whether Washington would ever exercise that authority is a political question; that it legally could is not in dispute.
Palantir has consistently contested that its European clients face meaningful exposure under the CLOUD Act, citing contractual protections and technical architectures designed to isolate data. Those assurances did not fully reassure the French parliament. In January 2026, the National Assembly voted to establish a formal inquiry commission on “structural dependencies and systemic vulnerabilities in the digital sector” — a signal that concern had reached a level of institutional seriousness requiring legislative investigation.
ChapsVision: a national champion — but an uncertain deployment
ChapsVision is a French company founded in 2019 by Olivier Dellenbach. It was approaching €200 million in annual revenue by end-2024, joined France’s Next40 index — the French government’s ranking of the country’s most promising technology scaleups — in June 2026, and has steadily built a presence in the European intelligence ecosystem through ArgonOS, its AI-powered data analytics platform designed to aggregate and cross-reference data from heterogeneous sources at scale.
The company had already won a first DGSI contract in 2024, focused on heterogeneous data processing. The competitive tender process, known as OTDH — launched in 2022 to identify a sovereign alternative to Palantir, with a program budget estimated at around €40 million — had identified three candidates: the Athea alliance led by defense contractors Thales and Eviden, a company called Blueway, and ChapsVision. The Athea alliance has since been described as a “dead letter.” ChapsVision remained.
But deploying ArgonOS inside the DGSI will take “a certain amount of time, probably several months,” and is expected to happen “probably in the course of 2027,” according to Economy Minister Roland Lescure. That timeline creates a double-run period with real financial consequences: France will be simultaneously paying Palantir under a contract that remains legally in force through 2028 and funding the ArgonOS deployment — the total cost of this overlap has not been disclosed. What is known is that the initial 2016 Palantir contract was valued at €10 million, and that the OTDH replacement program alone carries a budget of approximately €40 million. The two costs running in parallel, however briefly, represent a concrete price tag for a sovereignty decision taken six months after a renewal.
The French government confirmed that American tools will continue operating during the transition to avoid any “operational capability gap.” Replacing a data analytics platform inside an intelligence service is not like switching email providers. It requires migrating a decade’s worth of operational practice, retraining analysts, securing data flows during the overlap period when both systems run simultaneously, and verifying that the new solution performs under real operational conditions — meaning actual counterterrorism and counterespionage workloads that don’t pause for system migrations.
A European movement — and what shared architecture actually means
France is not alone. In mid-May 2026, Germany’s domestic intelligence service, the Bundesamt für Verfassungsschutz (BfV) — the German equivalent of the DGSI — announced it would also adopt ArgonOS to replace Palantir. The BfV had tested the platform over several months in counterterrorism and counterespionage scenarios before committing. BfV director Sinan Selen had signaled the direction months earlier: “We are well advised to sharpen our European focus.”
The significance of this convergence is technical as much as political. Because both the DGSI and the BfV are deploying the same platform — not two separate national solutions — the interoperability problem that might otherwise arise is structurally mitigated. ArgonOS is built on an open architecture: according to ChapsVision’s own technical documentation, the platform is designed to deploy “as a complete platform or as independent modules,” with “an open toolbox of components, connectors, and APIs” that allows integration with existing government systems. ChapsVision has also formalized this European ambition through partnerships with German IT integrators rola Security Solutions and SVA, specifically to deploy ArgonOS across European security authorities — with interoperability, the company states, as an explicit design guarantee.
This is a meaningful distinction from the scenario where France and Germany had each built their own incompatible sovereign tools. A shared European platform with open integration standards could, in principle, enable intelligence-sharing workflows that are both sovereign and interoperable. Whether that promise holds under the operational pressures of real-world intelligence cooperation remains to be tested.
Switzerland had moved earlier, having repeatedly declined to adopt Palantir solutions after a thorough assessment that concluded the risks of technological dependency and data access were “unacceptable.” In the United Kingdom, Palantir’s use in the National Health Service has come under renewed scrutiny.
Analysis
A history of unfulfilled promises
The Lecornu announcement is not the first time France has declared a break from American digital dependency. The Cloudwatt and Andromède projects, launched in the early 2010s to establish a sovereign French cloud infrastructure, ended in well-documented failure. In 2018, Laurent Nuñez — then director of the DGSI, now serving as Interior Minister — was already calling for the development of “a French or European alternative.” It took eight years and three contract renewals for that commitment to become an announced decision.
This precedent argues for measured expectations. The real measure of success will not be the June 2026 announcement, but the operational status of ArgonOS inside the DGSI by the end of 2027.
The power mechanics behind the reversal
How does a contract renewed in December 2025 get reversed six months later? Several factors converge. The geopolitical context has shifted: the Trump administration has heightened European governments’ perception of American unpredictability, making it politically untenable to depend on a company whose co-founder Peter Thiel — who also had early backing from the CIA’s venture capital arm, In-Q-Tel — maintains personal ties to the American president. ChapsVision’s technical maturity had also evolved: the company had already delivered a working DGSI contract in 2024, and Germany’s choice of ArgonOS weeks earlier had provided external validation of the platform’s readiness. And the April 2026 parliamentary report provided the documentary foundation to justify the decision politically.
The reversal is not irrational — it reflects a genuine convergence of factors. But the abruptness of the timeline, and the double-run cost it creates, suggest the decision was accelerated by political considerations as much as technical readiness.
What the CLOUD Act actually meant
The CLOUD Act is not an abstract risk. In practice, it allows U.S. prosecutors to obtain data held by American companies through a simplified legal procedure, without requiring the international legal assistance treaties — known as mutual legal assistance treaties, or MLATs — that traditionally govern cross-border data requests between democratic allies. The law was primarily designed for criminal investigations: drug trafficking, terrorism, cybercrime. But its scope is broad.
For the DGSI, this meant that a U.S. judicial order could theoretically have compelled Palantir to produce data processed on behalf of French intelligence — including information about active investigations, identified suspects, or potentially human sources. Whether Washington would ever have sought such an order is speculative; that it legally could have is not. For a counterintelligence service, that uncertainty is, structurally, intolerable.
The fundamental question: can you have intelligence sovereignty inside an alliance?
The deeper tension the DGSI-Palantir case exposes goes beyond ChapsVision’s deployment schedule. France belongs to NATO. It cooperates with allies in intelligence-sharing arrangements. And it has long used some of the same technical tools as its partners — including Palantir — precisely because that technical commonality facilitated cooperation.
The ArgonOS convergence between France and Germany offers a partial answer. If European allies migrate to a shared open-architecture platform rather than to separate national silos, the sovereignty gain need not come at the cost of interoperability. Think of it as the difference between every NATO country building its own proprietary radio frequency — which would make joint operations impossible — versus agreeing on common communication standards that each country implements with its own hardware. ArgonOS‘s modular, API-first design is at least an attempt at the latter approach. Whether it can sustain the data-sharing demands of real allied intelligence cooperation, at the classification levels involved, is a question that deployment will answer — not announcements.
The Bottom Line
ChapsVision will replace Palantir inside the DGSI — if the 2027 deployment timeline holds. The double-run cost is real but bounded. The interoperability risk is less severe than it appeared, because France and Germany are not building separate walls — they are, so far, moving toward the same platform. But the broader question this case raises remains open: whether France is building durable digital sovereignty, or simply transferring dependency from an American company to a French one. In ten years, if ArgonOS has become the indispensable supplier for the intelligence services of five or six European governments, the same questions will arise — concentration of supply, governance of source code access, the political difficulty of switching providers without operational disruption.
Digital sovereignty is not a destination — it is a discipline that has to be sustained, contract by contract, procurement by procurement, for as long as the stakes are this high.
Sources: French National Assembly Report n°2617 — Mission on France’s military dependencies vis-à-vis foreign powers (Cormier-Bouligeon / Saintoul, April 1, 2026) · DGSI official statements (2016–2026) · Prime Minister Lecornu speech, VivaTech, June 16, 2026 · Palantir press statement, June 16, 2026 · ChapsVision technical documentation (ArgonOS platform) · ChapsVision press releases (rola Security Solutions partnership, SVA partnership) · OpexNews · Theatrum Belli · France Info · AFP · Euronews · National Assembly Resolution n°2369 (parliamentary inquiry on digital dependencies, January 27, 2026)