Note: This article is a plain-language overview — not legal advice. For concrete implementation in your business, please consult a data protection officer or lawyer.
What it is actually about
As soon as your employees feed an AI with customer names, contract texts, or personnel files, this constitutes data processing. The GDPR applies — regardless of whether it is cloud AI or local AI. The difference lies in the effort required for compliance and the remaining risk.
Cloud AI: Processor Agreements and Third Countries
Processor Agreement (DPA)
You need a GDPR-compliant contract with the provider. This regulates, among other things: what is stored, who has access, for how long, what happens in the event of a data breach, and when is data deleted?
Third-Country Transfer (Schrems II)
Most large AI providers are based in the USA. Transfers there are legally sensitive following the ECJ ruling “Schrems II” — you must maintain standard contractual clauses plus risk analysis plus additional technical measures.
Practice
Even if a provider offers “EU hosting” — the parent company in the USA often has access via US jurisdiction (CLOUD Act). A completely secure GDPR statement is difficult to make here, even if many companies handle it pragmatically.
Local AI: Significantly fewer complications
If the AI runs on your own hardware and data does not leave your network, there is no external processor. There is no third country. The GDPR argumentation becomes simpler and, above all, honest.
But: Local systems must also be properly documented. “Local” is not a GDPR free pass.
What you must do even with local AI
- Expand the record of processing activities: which data, which purpose, which category of data subjects?
- Risk analysis: whether a data protection impact assessment is necessary depends on data and purpose.
- Employee instructions: clear rules of the game regarding what is allowed to go into the AI and what is not.
- Deletion concept: inputs and AI responses must be subject to a retention and deletion concept.
- Permissions: who is allowed to see what? RBAC per document and per use case — particularly important for RAG systems.
- Pseudonymization where possible — replace real names with IDs.
What we deliver
Every project includes templates for the record of processing activities, an employee guide, and a technical data protection concept. You do not have to start from scratch — and can pass the material on to your data protection officer or lawyer.
Local AI is not a GDPR free pass, but it makes compliance dramatically easier. Cloud AI can be used in a legally secure manner — but costs effort and trust in a corporation whose parent company is often based outside the EU.
GDPR-compliant AI setup in your business?
We include templates for the record of processing activities and a technical data protection concept.
Schedule an initial consultation
Frequently asked questions
Does the GDPR also apply to local AI?
What does third-country transfer mean regarding cloud AI?
Do I also have to document something with local AI?
Legal notice and privacy policy: Impressum & Datenschutz (German only; the German version is binding).