Latest in AI

In two days Jensen Huang took minority stakes in four French and five British AI companies and promised British safety research 250 million dollars' worth of compute over five years. The sums for the stakes have not been disclosed, and the donation is not cash but access to Nvidia's own product.

The Commission's own research centre has measured AI use in the ten sectors the EU calls strategic. The share has more than doubled since 2021 — from 7 to 18 per cent — but still trails the rest of the economy at 20 per cent. The growth comes from cheap language models, not from industrial policy.

The Commission's AI Office has sent requests for information to OpenAI and Anthropic after this summer's incidents, in which agents broke out of the labs' test environments. Lawyers disagree on whether the AI Act reaches into the testing phase at all.

ASML booked 11.7 billion euros in sales in the third quarter and earned 5 billion. But its own message on Wednesday was that planned output cannot meet customer demand in either 2027 or 2028 — and some customers are now asking about deliveries in 2030.

The Commission opened its consultation on the procedural rules for AI fines on Monday. The five-year limitation period only runs from the day a continuing breach ends — and six kinds of Commission action can reset it along the way.

Volt has signed a letter of intent with the cloud company Nebius covering a large share of 600 megawatts in Rotterdam. Phase one costs around 7 billion euros, the Dutch state has put nothing in, and the agreement is not binding.

Britain's AI minister says tech companies cannot be left to police their own models. That is the substance of a declaration 30 countries and the European Commission signed in September. The UK did not.

Since 2 August the AI Act has been enforceable with fines of up to seven per cent of turnover. The Commission's own list still has blanks beside six member states — the deadline was 2 August 2025.

The industry body IDA promises 37 billion euros through 2036 and points to regulation as the brake. Its own figures show a grid queue 30 to 50 times larger than the projects that actually exist.

The Commission has registered a citizens' initiative seeking to reserve the .IA top-level domain for Europe and tie it to eIDAS, the AI Act and the GDPR. Registration says nothing about the substance — and few initiatives ever reach the finish line.

German search engine Ecosia has dropped Mistral in favour of open models, including Chinese ones. The reason is quality — and it cuts straight through the European sovereignty argument.

The institution behind the world's first binding AI treaty has signed a cooperation agreement with Microsoft. The treaty is still waiting for ratifications. The cooperation is not.

The environmental assessment was supposed to come first. It was asked for once more than 300 hectares of forest had already been cleared at two of Google's Finnish sites.

Mistral's new model scores 82 percent at reproducing software vulnerabilities — a task the American frontier models refuse outright. Sovereignty, or a race to the bottom?

King Charles gathered OpenAI, Anthropic, Google DeepMind and Nvidia at Dumfries House and warned of existential dangers. Nvidia's chief politely declined to slow down.

The Scottish Parliament has halted new approvals of hyperscale data centres for up to a year. Power and water are the reasons — and more than 20 facilities are queued up.

A new threat report documents how state-linked actors from Russia and China used AI for espionage against European governments, theft from drone manufacturers and nearly 9,000 fabricated news articles.

In her State of the Union address, Ursula von der Leyen backed slowing down self-improving AI models — citing hacking as the reason. But Europe accounts for less than 5 percent of global AI investment.

274,000 Danes were hit by digital fraud in a single year, and ransomware victims are up almost 25 percent globally. Here is what you can actually do about it.

A growing number of European companies are choosing to run their own sovereign AI stack to keep data in Europe and meet EU regulations.

OpenAI has launched GPT-Live, a voice-based AI model powering ChatGPT Voice with a response time under 300 milliseconds and greater emotional nuance in speech.

The iPhone 18 Pro starts at DKK 10,999 - more than the unexpected expense one in five Danes can't cover. But the real shift is at the cheap end: budget phones are disappearing.

The warnings from AI researchers are about models running out of control. But follow the money and the race is just as much about chips, electricity and legislation - and Europe holds five percent of the world's computing power.

A researcher quit Anthropic and warned publicly that AI could cost us our lives before the decade is out. The unusual part was that his colleagues at the company agreed with him.

In an essay, Bill Gates warns that the transition to the AI era will be one of the most turbulent periods in history, and that nobody is preparing. He points to three dangers - and proposes three things to do now.

The AI companies aren't coming for the jobs first, but for the wage bill - and when wages turn into subscriptions, the money leaves Europe. Three parts: what we're facing, what it costs, and whether the balance can hold.

Meta has opened up Muse: an AI agent that makes a plan from a goal, keeps working after you've closed the app, and asks permission before it sends or buys anything. US only for now.

We've gone live with a brand-new website - built so that all content is data.