Palantir's European Exits and the Unresolved Question of Strategic Autonomy

France's Prime Minister Sébastien Lecornu did not lead with cost overruns or technical failure when he announced in June 2026 that the DGSI would end its relationship with Palantir in favour of French firm ChapsVision. He led with a question of control. France, he said, could not depend on partners capable of cutting access to critical technology at will, on their own terms, for their own reasons. The concern was not hypothetical. It had just been demonstrated. Weeks earlier, the US government had applied export controls to Anthropic's newest AI models, citing a cybersecurity vulnerability. Because the order took effect immediately, access was suspended for all users globally. European governments, researchers and enterprises found themselves locked out of tools they had been actively using, with no warning and no recourse. The controls were lifted eighteen days later. Not because Anthropic chose to cut off its own customers, but because operating from US soil meant its government could override its will regardless. The distinction between an unreliable company and an unreliable jurisdiction collapses the moment the company is subject to that jurisdiction. But the episode had done something that years of policy debate had not: it made the abstract concrete. The kill switch, long discussed as a theoretical risk in European technology circles, had been demonstrated in practice.

The instinct to frame this as either a political act or a technical one misses the point entirely. Whether the decision was driven by geopolitical calculation or a genuine safety concern matters less than the structure it exposed. A single foreign jurisdiction can unilaterally sever access, and no customer on the other end has any say in it. The reason the switch was flipped does not change what happens when it is and  it can happen again. Palantir is not the cause of Europe's anxiety. It is the most visible surface on which that anxiety has landed.

The pushback against Palantir did not begin with Fable 5. Switzerland rejected Palantir bids at least nine times on security grounds. The German debate over its use in state police forces had been running since at least 2023, when the Federal Constitutional Court issued a ruling setting strict fundamental rights standards for police data analysis that made Palantir's opaque architecture legally questionable. Lower Saxony's Interior Minister had already described it as "unmanageable" – not on performance grounds, but because the state could not audit or control what it was running. The Netherlands, Denmark and the UK had each signalled similar concerns. France's Health Data Hub had already chosen to replace Microsoft Azure with a French provider weeks before the DGSI announcement. Fable 5 did not create this trajectory, but it accelerated it. What June 2026 added was institutional architecture. The European Commission's Technological Sovereignty Package introduces a four-tier sovereignty framework for public sector procurement. The upper tiers require that providers are not subject to the jurisdiction of a third country – a structural exclusion that applies to every American technology company by definition, because every American company is subject to the US CLOUD Act, which grants US law enforcement authority to compel access to data regardless of where it physically sits. Palantir did not lose the DGSI contract because it failed. It lost it because of the legal jurisdiction it was born into.

And yet the picture that emerges from looking across European governments simultaneously is not one of coherent strategic withdrawal. While France announced it was exiting Palantir's civilian intelligence contract, NATO has been actively deepening its dependence on Palantir's Maven Smart System across its entire command structure. Acquired in March 2025 in one of the fastest sole-source procurements in alliance history, Maven achieved full operational capability on June 22nd 2026 certified for use on classified networks across exercises, missions and live operations. NATO's Task Force Maven director was careful to note that the system runs on NATO servers in NATO data centres with NATO-owned data. But the assurance reveals as much as it resolves. NATO owns the data. Palantir owns the system that processes it. Admiral Pierre Vandier, NATO's Supreme Allied Commander Transformation, acknowledged the bind directly: "As far as I know, today there is no real competitor for Palantir." His definition of sovereignty adjusted accordingly – in the short term, he argued, the realistic goal is not technological autonomy but control over data. France has begun developing its own competing military command system, Arcadia, with its Deputy Army Chief asking publicly whether the alliance should "adopt Maven blindly, or look for other solutions." French company Safran.AI, German firm Quantum Systems and UK company Hadean are actively building integrations on top of it — and Safran is simultaneously a partner in developing Arcadia, France's sovereign alternative to Maven. The same company is building into the system its own government wants to replace. Dutch State Secretary for Defence has said the government of Netherlands is working toward independence from Palantir within two years – while NATO's Joint Force Command Brunssum, located in the Netherlands, is one of Maven's early operational deployment sites.

The sovereignty argument is being applied where it is politically visible and operationally manageable, not where dependency is deepest or most consequential. Exiting Palantir from a domestic intelligence contract is difficult and will take years. Exiting it from NATO's warfighting infrastructure, if that ever became a political objective, would be a generational undertaking.

Meanwhile, Europe's private sector is not following its governments at all. A month before Lecornu's announcement, Thales signed a strategic cloud partnership with Google. Siemens uses AI models from the United States, Europe and China simultaneously. The European Commission's own assessment is unambiguous, the bloc depends on non-European providers for over 80% of key digital products and infrastructure. That number will not fall significantly in the near term, because procurement law applies to the public sector and the public sector is not where most of Europe's technology decisions are made.

There is also a structural trap no one has resolved. European alternatives to Palantir exist but cannot demonstrate operational capability at national scale without contracts and governments will not commit to contracts without demonstrated capability. One German provider made the comparison to the defence industry explicitly that what is needed are purchase guarantees, because without them the investment case for sovereign alternatives does not close. The Sovereignty Package mandates sovereignty before the sovereign alternatives exist at the required scale.

There is something worth pausing on about Palantir's own situation. Palantir is, by its own stated design, a company that only works with democracies. The European governments now pushing it out are precisely the kind of governments it built itself to serve. And its response to rejection, threatening legal action in London and objecting that "you can't do this on Instagram" in Paris, reveals a company that knows how to win contracts through technical superiority and legal leverage, but has not developed a political strategy for a world in which its most ideologically aligned customers have begun to treat American jurisdiction itself as the risk.

The most immediate thing worth watching is the DGSI transition to ChapsVision. It is the live test of whether sovereign alternatives can perform at national intelligence scale, and its outcome will matter well beyond France. A successful transition gives other European governments the evidence they need to follow; a troubled one gives their private sectors exactly the justification they need to keep choosing American providers regardless of what procurement law says. Companies like Thales, Siemens and Safran are currently running both strategies simultaneously – sovereign and non-sovereign  and ChapsVision's performance will determine which way that balance tips. Beyond that, two structural questions remain open: whether the Sovereignty Package's upper tiers survive contact with member states during negotiation without being commercially hollowed out, and whether Maven's growing operational depth inside NATO quietly normalises the very dependency that domestic procurement policy is trying to exit.

Fable 5 will not be the last demonstration that critical technology controlled from another country can be interrupted for reasons entirely outside state’s control. The next time it happens, the reason may be more political, the duration longer and the system more consequential. Europe understands this. What it has not yet resolved is whether understanding the problem is the same as being willing to pay the full cost of solving it.

_________________________________________

By Tamar Shengelia - Fellow of Cyber and Emerging Technologies Program at the Strategic Security Initiative (SSI)

Photo: OpenAI