Guide

Data Protection and AI Compliance and data protection: what the GDPR requires

As soon as personal data is processed in data protection and AI compliance, clear obligations apply. What these are specifically, what is almost always missing in practice, and what is actually sufficient.

Request assessment ↗Data protection and AI compliance

Quote upon request3 to 5 weeksStatus 2026

In short

As soon as personal data is processed in data protection and AI compliance, 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 point of collection, 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 data protection and AI compliance: a data processing agreement with each involved service provider, checking where processing takes place, a documented deletion period, and an understandable section in the privacy policy. Three things are almost always missing in practice: the defined deletion period, the notice at the point where the 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

May we enter customer data into ChatGPT?

In the free version without contractual safeguards: no. With a business account, data processing agreement, and training excluded, it may be permissible — depending on what data is involved. Precisely this distinction belongs in the usage policy, because it's constantly relevant in daily operations.

Do AI-generated texts need to be labeled?

For content that is published and synthetically generated or substantially modified, yes. Purely internal use — research, drafts, summaries — typically does not trigger this obligation.

Does this also apply to sole proprietorships?

Yes. The AI Act doesn't distinguish by company size, but by purpose and risk. Anyone with a chatbot on their website is affected — regardless of whether one or a hundred people work in the business.

How often does the inventory need to be updated?

Whenever a new tool is introduced or the purpose of an existing one changes. An annual review as a minimum interval makes sense, even without a specific trigger.

Do we need a data protection impact assessment?

Not generally. It becomes necessary when the processing is likely to pose a high risk to data subjects — such as extensive profiling. For an FAQ chatbot on a website, it's usually not required. We review this case by case and document the result in writing.

What if we use multiple AI tools from different providers?

Each tool is assessed individually — provider, data processing, and risk class differ. The usage policy summarizes the results so no one has to look up each tool separately.

Is a verbal agreement within the team sufficient instead of a written policy?

No, in case of dispute or audit, only what is documented counts. A brief written policy also protects new team members who never heard the verbal agreement.

05 — Next

Data protection and AI compliance by region

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.

Send inquiry ↗ office@pipeline-solutions.at OR +43 681 207 64 203