News - For Governments

An American browser just did more for Mistral's everyday reach than any EU programme has managed this year, while Spain logged its first breach run entirely by an AI agent and a new European cloud observatory gave institutional weight to a Broadcom dependency it has been tracking since March. Reach, detection and choice all moved today, just not toward the same owner.

Paris spent a day worried a Silicon Valley company might buy its AI champion, and Mistral denied it. Brussels' own cyber agency, meanwhile, had already run a security scan on EU code through an American model, and nobody remarked on it....

A report backed by Margrethe Vestager wants Europe to commit 100 billion euros to AI computing capacity or risk permanent foreign dependence. The funding that actually closed today ranged from a 350,000 euro data platform to a 300 million euro satellite manufacturer, useful money, none of it aimed at the specific gap the report describes....

Xi Jinping pitched an open-source AI coalition to BRICS that pointedly leaves the EU out, as Brussels' AI Act moves from paper to active fines. Sam Altman's OpenAI, meanwhile, wants the antitrust room to slow AI development that Europe closed off back in 2004.

Mistral's newly detailed 3. 5 billion dollar round rests on a single argument: enterprise control, not raw model power....

Mistral AI closed the largest equity round in European tech history even as a fresh survey found most European companies have quietly given up on full digital sovereignty. Meanwhile the Cyber Resilience Act's first hard reporting deadlines took effect against a backdrop of real breaches, and Brussels kept tightening its grip on Google and OpenAI under the DMA and AI Act.

A French AI model is joining a fifty satellite constellation worth a billion dollars, with government customers in France, across Europe and the UAE. The same day, regulators in France, Germany, Spain and Norway kept finding the data protection failures that Brussels' own new deal with Washington risks repeating at a larger scale.

ENISA can now put Anthropic's Mythos 5 and OpenAI's GPT-6 Astra through real testing under the EU AI Act, oversight a new industry survey calls ‘resilient interdependence’ rather than sovereignty. Bending Spoons closed a 1....

Bern is piloting a homegrown desktop suite for three thousand civil servants, and a European space firm just banked $450 million for reusable rockets, real infrastructure moving while regulators write labels beside it. A US legal AI unicorn opened a Dublin office off the back of a $550 million round the same week, and hired the payroll without moving an inch of control....

Mistral just closed Europe's largest ever tech funding round to build the compute layer this column keeps saying the continent talks about more than it builds, except one of its two lead investors isn't European. The same day, Google's EU search overhaul shows what it actually costs when Brussels' rules bite instead of just existing.

Thousands of OpenAI's agents spent weeks running loose on a dormant German wiki, and a Heise report surfaced it before an Article 55 incident filing followed. The same day, Brussels' own fix for online age checks draws exactly the privacy criticism the platforms it regulates have spent two years failing to answer.

Six days before the Cyber Resilience Act forces companies to report exploited flaws within 24 hours, Berlin found out that refusing a ransom does not stop stolen data from surfacing anyway. The same week, Europe's cloud lobby accuses Broadcom of widening a fight it started over VMware licensing into control of the enterprise AI layer.

Brussels tightened its grip on ChatGPT, Reddit and Roblox this week while quietly rewriting parts of its own privacy rulebook. Underneath the regulatory noise, Europe's sovereignty ambitions kept colliding with market realities, from Italian cloud tenders that reward US hyperscalers to Nvidia's purchase of a French AI icon.

Rome's public procurement rules for cloud services were designed to keep European capacity in the running, and instead they hand the field to American hyperscalers by default. The same day, Google's model for who gets to use its fastest vulnerability-hunting AI shows a second kind of sovereignty gap: not a rule that misfires, but one that was never written.

What this column called a rumour on 27 August has a signature now: Nvidia has confirmed it is buying Hugging Face, the French-founded hub much of Europe's open-source AI runs on, for $12. 93 billion....

The EU's General Court told Opera that Microsoft Edge isn't a gatekeeper on the same day the Commission opened a licensing probe into Oracle's cloud business, a split verdict that says more about where Brussels sees real market power than any single ruling could. Meanwhile a German spacetech raise shows sovereign infrastructure money moving at last, not just sovereign-AI branding.

The day Brussels' Digital Omnibus sends its rewrite of what counts as personal data into contested interinstitutional negotiation, Austria's Supreme Court just enforced the GDPR principle at stake against a credit agency that broke it. Meanwhile two European firms show what closing the sovereignty gap by building, rather than legislating, actually looks like.

The European Commission has designated ChatGPT a Very Large Online Search Engine under the Digital Services Act, the first time an AI chatbot has been slotted into that legal box, alongside Reddit and Roblox as very large online platforms. All three now have a four-month window, running out around January 2027, to show they can contain systemic risk, the same day the Netherlands published exactly the kind of infrastructure alternative that rulebook was always meant to protect.

Europe spent the week confronting how much of its digital and physical infrastructure it does not actually control, from satellite timekeeping to police databases shared with Washington. Regulators answered with some of their sharpest enforcement yet, while investors poured fresh capital into robotics and deep tech built to reduce that dependence....

A historian's warning about an Artificial State, where algorithms quietly absorb public functions, landed the same week Europe got a far more literal version of the argument: GPS jamming and spoofing serious enough to worry the European Space Agency. Add OpenAI formalising a policy-only role for Dublin and Brussels while it hires for commercial growth everywhere else, and today reads like three case studies of the same dependency, told at three different scales.