PIPELINEGuide
AI training and data protection: what GDPR requires
As soon as personal data is processed in AI training, clear obligations apply. What these are specifically, what's almost always missing in practice, and what actually suffices.
As soon as personal data is processed in AI training, you need a legal basis, purpose limitation, defined deletion periods, information for data subjects, and a data processing agreement with every service provider involved. In practice, the same three points are almost always missing: the deletion period, the notice at the collection point, and clarification of whether the provider uses the data for its own training.
01 — Data Protection
What GDPR requires
As soon as personal data is processed — and that happens faster than expected, for instance as soon as someone enters a name or phone number — the usual obligations apply: name the legal basis, define the purpose, maintain data minimization, define retention periods, inform data subjects.
In practice, this means for AI training: a data processing agreement with every service provider involved, checking where processing takes place, a documented deletion period, and a clear section in the privacy policy. Three things are almost always missing in practice: the defined deletion period, the notice at the point where data is collected, and clarification of whether the provider uses the data for its own training.
A data protection impact assessment is only required when there's likely high risk — not for most projects of this type. The review itself should still be documented so that if questions arise, it's clear why it was omitted.
02 — Contract
What needs to be clarified in writing
- Ownership of data and access. They belong to the client. If a provider won't clearly confirm this, that's the most important information from the conversation.
- Handover upon termination. In what format, within what timeframe, at what cost.
- No minimum term without consideration. Lock-in is justifiable if something is delivered in return — otherwise not.
- What a later change costs. The total cost isn't determined by the project price, but by the hourly rate in year three.
- Documentation as a deliverable. Not as a promise, but as a line item in the proposal.
03 — Risks
What can go wrong
- The system answers incorrectly. Containable by having it work exclusively from verified content and escalate when uncertain rather than guess. This is a configuration decision, not a property of the technology.
- The effort is underestimated. Most common reason: unclear starting situation. An honest assessment of the current state before the proposal prevents this more reliably than any buffer.
- No one maintains it. The biggest risk of all — and the most easily avoidable.
- Data protection gaps. Almost always arise from missing deletion deadlines and unclear responsibility, rarely from the technology itself.
- Vendor lock-in. Avoidable if access credentials, data, and documentation contractually remain with the client.
04 — Questions
Frequently asked questions
Do you also cover industry-specific applications?
Yes, as far as known in advance. Bring your typical tasks — quote texts, protocols, customer correspondence — and we'll practice on exactly those instead of generic examples.
Does the EU AI Act require training?
For businesses that use AI systems, there is an obligation to ensure adequate AI competency among the people working with them. A documented training is the easiest way to demonstrate compliance.
Can we do this online as well?
Yes, then usually in two half-days instead of one full day — concentrated screen work over eight hours doesn't work well.
Do our people need prior knowledge?
No. We start from zero and divide the exercises according to prior knowledge. Experience shows that every team has both extremes.
What if individual employees don't want to?
That happens and it's not a disaster. We train everyone equally, but no one is forced to use it — experience shows that changes on its own once colleagues see the benefits.
Is this also useful for executives who don't do much typing themselves?
Yes, for a different reason: They need to be able to assess where these tools make sense in the team and where risks lie — that's independent of whether they work with them daily themselves.
Won't that be outdated after three months?
The tools change, the mindset doesn't. The focus is on when such a system is appropriate and how you verify its output — that remains valid even if the interface looks different.
Next step
Initial assessment, no obligation.
Briefly describe what you need. You'll get an honest answer about what's realistic and what it will roughly cost.