AI in exams: how lexilink meets Hamburg's HIBB guideline
Hamburg's vocational schools now have binding rules for AI in assessments. Here's how lexilink is built to meet them.
Why this matters
Since 17 April 2024, the Hamburg Institute for Vocational Education (HIBB) has enforced a binding guideline on the use of Artificial Intelligence in connection with performance assessments and examinations. It applies to every programme at Hamburg's vocational schools — and to every teacher who uses AI to create or run an assessment.
lexilink uses AI. So we asked ourselves: what does a learning tool have to do so that teachers can meet this guideline without extra work? This post explains how we align lexilink with it.
What the guideline requires
The guideline rests on one principle: schoolwork and exams must be the student's own, independent work. Five requirements follow:
- The teacher decides (§2) for which assessments AI may be used — and tells students the rules in advance.
- Transparency (§3.1): if AI is used, the tool, version, URL, prompt, date, output and form of use must be documented — a *Hilfsmittelverzeichnis* (record of aids). Format:
<Prompt> prompted by <First Last>, date, URL. - Data protection (§3.2): no personal data into AI tools; no uploading documents from which individuals could be identified.
- Copyright (§3.3): AI often reproduces protected content unmarked; reusing it unchecked infringes copyright.
- Academic integrity (§4): undisclosed AI use counts as cheating (Täuschung). A declaration of independent work is recommended.
Where lexilink uses AI — and where it doesn't
This is the key distinction: in lexilink, AI helps the teacher create tasks, content and practice questions. Student work is not graded by AI. Closed questions are scored deterministically; open questions are graded by the teacher — as always.
No black-box grading. No AI deciding pass or fail. Pedagogical responsibility stays where it belongs.
How lexilink meets each requirement
§2 — The teacher decides. For every test, the teacher sets an AI policy: *forbidden*, *permitted with disclosure*, or *permitted*. Students see that rule before the test begins.
§3.1 — Transparency in one click. When a teacher authors tasks with AI assistance, lexilink automatically logs the tool, model, prompt and date — and generates a Hilfsmittelverzeichnis in the prescribed format, exportable as evidence. What the guideline mandates as a duty becomes a feature.
§3.2 — Data protection and data sovereignty. lexilink is built to run AI requests through its own language model, hosted in the EU — without handing inputs to US providers like OpenAI. We're completing the production rollout of this EU hosting as part of this initiative. lexilink also warns when personal data is detected in input fields.
§3.3 — Copyright. When generative features are used, lexilink prompts careful handling of protected content and raises awareness among teachers and students.
§4 — Independent work. Before submitting, students affirm a declaration of independent work whose wording follows the test's AI policy. Rule-based integrity signals support the teacher's judgement — without AI surveillance or automatic suspicion.
Our stance
AI is a tool, not a substitute for learning. Good tools don't make the rules harder to follow — they make them easier. That's why we build transparency, teacher control and EU data sovereignty straight into the product, instead of offloading them onto teachers as a checklist.
In the interest of transparency
Aligning lexilink with this guideline is an ongoing effort. Features such as the per-test AI policy, the automatic Hilfsmittelverzeichnis and the declaration of independent work are rolling out as part of this initiative. We share the current status openly.
For schools and institutions
Do you work at a vocational school in Hamburg or beyond and want to use lexilink in a privacy-compliant way? Get in touch for a compliance overview or a pilot.