Denmark made students defend their essays out loud. It's the first national admission that AI broke written homework.

Denmark's Education Ministry has ordered every upper-secondary student who writes a take-home essay to defend it orally in front of a teacher. The policy applies to roughly 9,000 students aged 16 to 19 and takes effect immediately as pupils return to school next week. 1 2

Heunicke called it an emergency response. "Unfortunately, we have a problem with students using AI to cheat in upper secondary school. Action is needed now, and we are starting with these three initiatives." 1 Read structurally, the measures are an admission that no software can solve this problem. When you cannot verify who wrote a document, the only reliable test of authorship is whether the student can stand up and explain it.

What Denmark mandated

The three measures take effect immediately:

• Oral defenses for all written exams completed at home, including the SSO (større skriftlige opgave), the major annual written assignment that about 9,000 upper-secondary students complete each year. 2
• Screen-monitoring tools to oversee how computers are used during in-school written exams, plus firewalls to filter what is available on school networks during both exams and classroom teaching. 3
• A push to move more assignments onto campus under teacher supervision, where the ministry says teachers can "follow the students' work process, provide relevant feedback and create a more accurate basis for assessment." 2

Three organizations representing school leaders, teachers, and students welcomed the move but warned it is only a stopgap given the "rapid pace of technological development," according to Danske Gymnasier, the high schools association. 1 Anders Frikke, chair of the Danish Association of Upper Secondary School Teachers (GL), said teachers want additional steps: requiring students to declare when they have used AI in assignments and ensuring oral exam preparation happens without AI access. 1

Why detection was already losing

The policy bypasses an entire product category. Since ChatGPT's launch, schools worldwide have leaned on AI-detection tools from companies including Turnitin, GPTZero, and Copyleaks, which advertise accuracy rates above 90%. 4 But peer-reviewed research and institutional reviews have repeatedly shown these tools produce false positives and are particularly prone to flagging work by non-native English speakers and neurodivergent students. 5 6

That creates a trap schools cannot escape at the text level. Detect too aggressively and you wrongly accuse honest students. Detect too leniently and AI-generated work passes through. Denmark's answer is to stop reading the text for telltale signs and start testing the student directly.

The scale of the problem next door shows why the urgency is real. In neighbouring Sweden, the number of students suspended or warned for unauthorized AI use rose 688% between 2023 and 2025. 2 The Swedish government has commissioned a review of how schools handle AI cheating, due to report back in April. 2 Denmark did not wait for a study.

Three countries, three bets on the same problem

The Nordic region has become a natural experiment in how education systems respond when AI breaks written assessment. Within months, three neighboring countries placed structurally different bets:

• Norway imposed a near-ban on generative AI for elementary school pupils and restricted its use for older students. 3
• Sweden commissioned a government review after suspensions surged 688% in two years. 2
• Denmark redesigned assessment itself, replacing a text-level integrity check with a person-level one. 1

Denmark's bet concedes the most. Norway tried to restrict access. Sweden tried to study the problem. Denmark accepted that the medium itself is compromised and changed what it evaluates. And it is the same country that, just last year, ran an experiment allowing students to use AI during English oral tests in upper-secondary schools. 3

That pivot is the signal every school board, university, and edtech investor should watch. If the oral-defense model spreads beyond Denmark, the market for AI-detection tools built on the premise that software can separate human from machine writing loses its clearest policy-level use case. The technology will not disappear. But a national government just said, in practice, that it cannot be trusted to adjudicate who did the work. When that conclusion reaches higher education and corporate training, the question will be the same one Denmark just answered: can the person explain it?