The world's number-one digital government is about to run on AI.

Danes meet their state through one shared digital infrastructure: identity, mail, tax, health, welfare. Analogue access survives only as an exemption: paper post for the few, no parallel administration. Denmark ranks first of 193 in the UN's e-government index, and the government expects AI to be able to free at least 30,000 full-time work-years across the public sector. In a state this connected, everything spreads at once, benefits and mistakes alike. Getting Denmark right is cheap relative to what it teaches: a playbook for every digital democracy that follows, in a union where rules often begin as one member state's practice.

One example of what "getting it right" means. Used well, AI in public administration means faster answers and better service for citizens, and the government plans a general legal basis to get there. But the safeguards EU law promises when a machine decides a case outright (human review, a hearing, a right to contest) do not attach automatically on that legal track, and no Danish law gives a citizen a general right to have a human redo a machine's decision. The gains and the safeguards have to be built together.

The second half of the case is political. Denmark has no independent institution working full-time on AI policy, and that vacancy is the opening for a middle power to do more than deploy AI well: drive the sovereignty agenda at home and in the EU, and become a champion of AGI and ASI governance internationally. The UK shows what sustained briefing and advocacy can build: more than 125 cross-party parliamentarians have signed a statement, organised by the advocacy group ControlAI, calling for binding regulation of the most powerful AI systems.

This is societal adaptation: a small state cannot decide what gets built in San Francisco or Beijing, but it decides its own preparedness, whether humans stay accountable inside human-AI systems, and whether middle powers act as a counterweight when great powers hesitate to restrain each other.

The technical ground is shifting under every government at once. The UK's AI Security Institute has measured autonomous cyber capability doubling roughly every five months, and in July OpenAI reported a frontier model breaking out of an isolated test environment and compromising real infrastructure. Alignment, biological misuse, autonomous cyber operations, AI in weapon systems: the core safety problems remain unsolved while capability compounds.

The Danish conversation has not caught up. FRIKTION has mapped it: of the more than 3,500 AI-related newspaper items we counted across five national newspapers from January to July, less than one per cent had safety as the main frame. Coverage runs on exams, deepfakes and fraud; on AI-enabled biological misuse, the count of substantial pieces was zero. Danish journalists catch the big events; what no institution supplies is the sustained depth that turns attention into judgement. Raising AI literacy in the country that automates first makes every other decision better.

The state is not yet organised for it either. The government has set a strategic direction for AI, but no integrated plan connects adoption, competitiveness, infrastructure, frontier security and the labour market, with priorities, budgets and clear owners. The bill that would have completed Denmark's AI supervision fell with the spring 2026 election, and big infrastructure choices, such as whether new datacentres can connect to the power grid, are being made one at a time rather than as parts of one plan.