AI data privacy: which infrastructure choices help?
Which infrastructure choices help protect privacy with AI: region, encryption, access, logging and contracts.
Protect data in AI with data location within the desired region, encryption at rest and in transit, strict access rights, limited logging and clear agreements with providers on the use, retention and deletion of data. Classify your data first, so you know which data does not belong in a public AI service.
Who is it for: Organisations that want to use AI with personal data or confidential company data.
Know which data you use
Start with an overview of the data that ends up in AI applications: personal data, customer data, trade secrets, public information. That determines which protection is needed and which services you may use. Sensitive data deserves a stricter environment.
Choose region and location
Look at where data and models are processed and stored, including backups and logs. For personal data, the GDPR may impose requirements on transfers outside the European Economic Area. Also read the guide on data residency and ask providers where support and management take place.
Technical measures
- Encrypt data during storage and transmission and decide who manages the keys.
- Grant access according to the principle of least privilege.
- Separate environments for development, testing and production.
- Limit and protect logging, because input can contain personal data.
- Consider omitting or replacing personal data before data goes into a model.
- Set retention periods and delete what is no longer needed.
Agreements with providers
- Sign a data processing agreement when a provider processes personal data for you.
- Ask whether input is used to train models and whether you can switch that off.
- Check subcontractors and processing locations.
- Record what happens to your data after termination.
Policy and oversight
Record which AI applications are permitted and which data may go into them. For processing that is likely to result in a high risk, carry out a data protection impact assessment (DPIA, Art. 35 GDPR). Keep an eye on developments in the European AI Act and seek legal advice where necessary.
Measures and what they protect
| Measure | What it protects | Note |
|---|---|---|
| Region choice | Legislation and transfers | Include backups and management too |
| Encryption | Data at rest and in transit | Key management determines the protection |
| Access management | Misuse and leaks | Check permissions regularly |
| Limited logging | Sensitive input | Logs also have retention periods |
| Contracts | Use and deletion by provider | Check subcontractors |
Who is it not ideal for?
- Anyone who processes only public, non-sensitive information.
Frequently asked questions
May I paste personal data into a public AI service?
That depends on the agreements, the legal basis and the provider. If in doubt, ask your data protection officer or a lawyer, and set a policy.
Does a dedicated environment help automatically?
It gives more control, but you are yourself responsible for security and compliance.
Do I have to report anything in the event of a data breach?
Possibly. In the event of a breach involving personal data, consult your data protection officer or legal adviser straight away.
