Denmark is piloting free AI access in English oral exam preparation from 2026, allowing students to use ChatGPT and similar tools during the hour before they face examiners. A new experiment reveals what happens when you give teenagers unrestricted AI help and how they actually use it.
The Danish education ministry has launched a closely watched experiment. Starting with this summer’s exams, volunteer upper secondary schools will let students use generative AI tools during the one hour preparation phase ahead of their English oral exam. The exam itself remains analogue and tool free. But that prep hour is now wide open.
I have watched Denmark tinker with digital exams since internet access was allowed back in 2008. This feels different. The tools are more powerful. The stakes are higher. And nobody quite knows what will happen when you hand a stressed 18 year old a chatbot and say go ahead.
What Students Actually Did With Free AI Access
According to a report by DR, the first batch of results from participating schools shows a pattern. Most students used AI for language checking, idea generation and structuring their notes. Some asked it to rephrase awkward sentences or suggest better vocabulary. A smaller group went further and asked the AI to draft entire speaking scripts or summarise complex texts they had not read.
Teachers could not monitor every query in real time. They relied on the oral performance itself to catch students who leaned too heavily on AI generated content. If you cannot explain what you brought in, examiners notice. Grades can drop accordingly.
The ministry frames this as modernisation. Students will meet AI in almost any future workplace, officials argue, so schools should teach them to use it critically rather than pretend it does not exist. The setup tries to split the difference. AI for prep, human performance for assessment.
The Legal and Practical Tightrope
Denmark is not winging this. A new guide published in January 2026 by the Ministry of Children and Education lays out how schools must handle AI lawfully. Institutions must ensure tools comply with GDPR and the incoming EU AI Act. They bear the responsibility, not the tech companies or individual teachers.
That means checking whether student data flows to servers outside the EU. It means vetting tools for bias and reliability. It means understanding copyright when AI generates text or images. Many schools are steering students toward vetted platforms like Microsoft Copilot within Office 365 rather than letting them loose on open web chatbots.
The legal emphasis is deliberate. Denmark has long pursued digital innovation in the public sector, but generative AI raised fresh compliance headaches. The guide tries to shift the burden from confused teachers to institutions with legal departments. Whether smaller schools can actually meet those standards is another question.
A Split Strategy on Written Versus Oral Exams
While opening the door for AI in oral prep, the ministry has simultaneously tightened the rules for written English exams. Parts of those tests must now be completed by hand, reversing years of liberalisation. The logic is that written tasks are more vulnerable to invisible AI delegation. Oral exams test spontaneous communication, which is harder to fake even with a polished script in hand.
This divergence confuses some students and worries equity minded educators. Those already skilled at prompting AI and evaluating its output gain an advantage. Students from resource strong homes have had months or years of informal practice. Others rely solely on what schools provide, which varies wildly.
Sparring or Self Deception
A widely cited column in Computerworld draws a sharp line between using AI for sparring and using it for delegation. Sparring means bouncing ideas off the bot, asking it to check grammar or suggest alternatives. Delegation means handing over the task and accepting whatever comes back.
Most Danish companies, the piece argues, still only use AI for simple text help. Emails, meeting summaries, slide decks. The same likely applies to students. But the columnist warns that people often confuse the two and convince themselves they are being productive when they are really just outsourcing thinking.
I see that risk in classrooms. A student who uses AI to structure an argument might learn something about rhetoric. A student who pastes in a prompt and copies the output learns only how to copy and paste. The one hour exam prep window makes it very hard for teachers to tell the difference until the student starts speaking.
What This Means for Learning
Educators are split. Some believe AI can support weaker students, offering scaffolding that helps them keep up and participate more confidently. Others fear it will mask real literacy problems and erode the habit of deep reading and original thought.
There is almost no robust Danish data yet on learning outcomes under AI assisted conditions. The English oral pilot is supposed to generate some, but evaluation details remain vague. Meanwhile, students are already using ChatGPT and similar tools informally across subjects, making any controlled experiment somewhat artificial.
Language teachers raise a particular concern. If students lean on AI for polished phrasing during prep, they may sound more fluent on paper than they are in spontaneous speech. The oral exam itself should reveal that gap, but only if examiners are alert and tasks are designed to test real time thinking rather than recitation.
International Eyes on a Danish Experiment
Denmark is not alone in wrestling with AI in education, but it is among the first to formalise access in high stakes national exams. Other European countries initially responded to ChatGPT with bans or tight restrictions. Denmark has generally avoided blanket prohibitions, relying instead on guidance and local discretion.
That approach reflects both the country’s longstanding trust in digital tools and its tolerance for experimentation. It also reflects a smaller, more cohesive education system where policy changes can be piloted and adjusted relatively quickly. International bodies like the OECD are watching closely. If the Danish model works, others may follow. If it creates chaos or widens inequality, expect a sharper regulatory backlash.
For now, the experiment is live. Students are using AI. Teachers are learning on the job. And the rest of us are waiting to see whether free access to generative tools during exam prep turns out to be a pragmatic step toward AI literacy or a shortcut that hollows out the very skills exams are supposed to measure.
Sources and References
