AI and student feedback: the privacy checklist for your data protection officer

Open answers are personal data as soon as they can be traced to a person, and free text contains something traceable more often than you'd think: a name, a teacher, a situation. That is why it pays to have it settled beforehand.

The ten questions

1. Where is the data processed? Within the EU is simplest. Also ask where the AI models run, not only the database.
2. Are the answers used to train AI models? The answer should be a clear no, and it should be in the contract.
3. Is there a data processing agreement? A vendor who processes personal data on your behalf is a processor. That calls for a data processing agreement (Article 28 GDPR).
4. Which subprocessors are used? Think of the hosting, the AI provider and the mail. Ask for a current list and how you are told about changes.
5. Who owns the data? The institution should remain the owner, with the ability to request everything and have it deleted.
6. How long is data kept? Ask about the default retention period and whether you can set it yourself.
7. Can answers really be anonymous? Ask what anonymous means: does the vendor or your organisation see who said what? And what happens in a small group, where an answer can be recognisable by its content?
8. Do students know they're talking to an AI? They should: the EU AI Act asks for transparency when people communicate with an AI.
9. Can you see where a conclusion comes from? A summary without a source can't be checked. Every theme should come with the count and the verbatim quotes.
10. What happens when something goes wrong? Ask about the data breach procedure and the security measures.

Three things that are often forgotten

• What students type themselves. In an open field someone can write about a health situation or a private problem. That is sensitive data, even if you didn't ask for it. Think beforehand about how you handle it.
• Teachers appear in it too. Students write about teachers, often by name. That is personal data about those teachers. Agree who sees what.
• A DPIA may be needed. With new technology and large-scale processing a data protection impact assessment (DPIA) may be in order. Your DPO knows whether that applies here.

What about your own AI licence?

Some universities already work with their own AI licence through their cloud contract, where the data isn't shared. Others wait for policy. Both can work, as long as you know where the data goes and who can reach it. A personal account of an employee for work data is in any case not a good route.

How Febaso does it

• Processing in the EU, and never used to train AI models.
• A data processing agreement with a list of every subprocessor. The institution stays the owner of the data, and data is deleted on request.
• Anonymous when you choose: you choose per interview whether answers are anonymous, and then you never see who said what.
• Open about the AI: students know from the first message that they are talking to an AI.
• Every theme with the count and the verbatim quotes, so it can be checked.

More about how we handle privacy is on the privacy page. For your DPO we can provide a summary; ask for it in an intro.