Teaching AI Under New Rules: Adaptations and Frustrations with the EU AI Act

Sunday Reflection — 2026-10-18

I spent ten minutes last Thursday watching six smart students argue with a risk assessment form. Not about the ethics of their AI project — but how to check the right boxes on a document none of them had seen before. Welcome to teaching AI under the EU AI Act: full of noble aims, plenty of awkward reality. The Act promises safety and transparency for everyone building AI. In the classroom, it mainly promises extra paperwork before anyone can write a line of code.

mermaid flowchart TD A[Student proposes project] --> B[Mandatory risk assessment] B --> C{Is risk assessment approved?} C -- Yes --> D[Proceed with model deployment] C -- No --> E[Revise proposal, repeat review] D --> F[Project work begins] E --> B

Last month a group of career-changers tried to build a chatbot for crisis counseling, a good fit for Berlin. But before they could test it, the school’s compliance officer flagged their questionnaire as “potentially high-risk”. Suddenly, what should have been a week of hands-on work turned into three weeks of emails and amendments. In the end, they switched to a less sensitive, far duller project—more likely to hit the right regulatory boxes but far less likely to teach them real world skills. I’ve seen projects die in meetings before, but rarely at this speed.

There are benefits. Students now have to think harder about safety and unintended consequences. I wouldn’t want to go back to the days where someone could throw a black-box model onto a live site just to see what happened. But in practice, the rules freeze a lot of ambition and slow the best learning moments: the bit where you see your code break in the wild, and figure out why. What stands out is the difference in reaction. Germans mostly accept the new forms as just one more process; it’s the British students, trained on “try it and see”, who get frustrated having to ask for permission first. My job ends up somewhere between interpreter and negotiator.

Maybe this is another one of those things that works as a principle but stumbles in the day-to-day. You can improve safety without killing curiosity—but, as of now, the balance is off.