Practical guide
Using AI in compliance with data protection law: a guide in seven steps
Clarify purposes, check vendors, document risks and involve your team
Using AI in compliance with data protection law means: you know which data goes into which tool, you have a legal basis, a data processing agreement and clear rules for your team, and you can provide information to data subjects. This guide takes you in seven steps from taking stock to day-to-day operation. It does not replace legal advice; discuss the results with your data protection officer.
Before you start
- Assign responsibility: Name a person to coordinate the rollout and involve your data protection officer from the start.
- Take stock: Find out which AI tools are already in use in the company, including services set up privately. Whatever is used without approval is called shadow AI. Why it is a risk is explained in our article Shadow AI: What matters when choosing an AI provider.
- Involve the works council: If there is a works council, bring it in early. Technical systems intended to monitor the behaviour or performance of employees are subject to its co-determination rights (Section 87(1) no. 6 of the German Works Constitution Act, BetrVG).
1. Define the purposes
For each use, describe what the AI is used for: for example questions about internal manuals, drafts of quotes or a chatbot on the website. A clearly defined purpose is a prerequisite for purpose limitation under Art. 5(1)(b) GDPR and makes every further step easier. A goal such as "AI for everyone" can neither be assessed nor checked.
2. Classify the data
For each use, clarify which data the AI processes.
| Type of data | Examples | What follows |
|---|---|---|
| No personal data | Product data sheets, published price lists | The GDPR does not apply; still protect trade secrets |
| Personal data | Names and contact details in emails, quotes, minutes | Legal basis, information for data subjects, data processing agreement |
| Special categories | Health information, trade union membership | Additionally an exception under Art. 9(2) GDPR; check whether a data protection impact assessment is needed |
| Employee data | Performance data, personnel files | Check co-determination, restrict access tightly |
3. Determine the legal basis
Every processing of personal data needs a legal basis under Art. 6(1) GDPR. In companies, this is usually the performance of a contract (point (b)), for example when answering a customer enquiry, or legitimate interests (point (f)) with a documented balancing test. Special categories additionally require an exception under Art. 9(2) GDPR, often explicit consent. Record the legal basis in the processing register under Art. 30 GDPR.
Public authorities usually rely on the performance of a task carried out in the public interest (Art. 6(1)(e) GDPR) in conjunction with a statutory basis. What the public sector should also consider is described in our article AI in the public sector.
4. Check the vendor
Most AI tools run with an external provider. The provider should answer these questions in writing:
| Question for the provider | What to look for |
|---|---|
| Is there a data processing agreement under Art. 28 GDPR? | Complete, with technical and organisational measures and the list of subprocessors |
| Where does each processing step take place? | Storage, search, answer generation, text recognition and transcription named individually |
| Is data processed outside the EU? | A basis under Chapter V GDPR, such as an adequacy decision or standard contractual clauses with an assessment of the level of protection |
| Are inputs or documents used for training? | Contractually excluded or demonstrably switched off |
| How is data deleted? | Retention periods, deletion on request, export at the end of the contract |
| How are permissions enforced? | The AI only shows what the person is allowed to see anyway |
How to recognise a transparent provider is covered in more depth in our article AI providers and data protection.
5. Assess and document risks
- Data protection impact assessment: Under Art. 35 GDPR it is mandatory if processing is likely to result in a high risk to data subjects. The list of the German Data Protection Conference names, under number 11, the use of artificial intelligence to steer interaction with data subjects or to evaluate personal aspects, for example in customer support.
- Update the processing register: Enter each use with its purpose, types of data, recipients and retention periods in the processing register under Art. 30 GDPR.
- Check the AI Act: If a use falls under Annex III of the AI Act, such as recruitment, it is a high-risk system; following the Digital Omnibus, the obligations for it apply from 2 December 2027. Since 2 August 2026, chatbots have been subject to the disclosure obligation under Art. 50; more on this in our article GDPR-compliant chatbot.
6. Rules and training for your team
- AI policy: Define which tools are approved, which data must not go into them, such as personnel files or health data, how results are to be checked and who helps with questions.
- AI literacy: Under Art. 4 of the AI Act, companies that provide AI systems or use them professionally take measures to support the development of AI literacy among their staff. Since the amendment by the Digital Omnibus on 27 July 2026, they do not have to guarantee a specific level of literacy. Document training and guidance anyway.
- Check the results: AI answers are drafts. Amounts, deadlines and legal statements are confirmed by a person before being passed on.
7. Operation: rights, deletion, control
- Data subject rights: Access under Art. 15 and erasure under Art. 17 GDPR must also work for chat histories and AI logs.
- Retention periods: Only store data for as long as the purpose requires (Art. 5(1)(e) GDPR).
- Permissions: Regularly check who can access which knowledge areas, especially after changes in the team.
- Changes: Assess new features and integrations before you enable them.
Checklist
| Step | Result | Evidence |
|---|---|---|
| Before you start | Responsibility, inventory and works council clarified | List of tools in use |
| 1. Purposes | Purpose described for each use | Description per use |
| 2. Data | Types of data classified for each use | Data type table |
| 3. Legal basis | Basis determined for each use | processing register under Art. 30 |
| 4. Vendor | Questions answered in writing | Data processing agreement, list of subprocessors |
| 5. Risks | DPIA and AI Act checked | DPIA or reasoned note |
| 6. Team | Policy and training | AI policy, training records |
| 7. Operation | Rights, deletion and permissions arranged | Deletion concept, permissions overview |
How Nexoria supports your review
For your data protection review, the data processing agreement under Art. 28, the processing register under Art. 30 and the data protection impact assessment under Art. 35 GDPR are publicly available: DPA, processing register, DPIA. The details on this page are Nexoria's own statements; the data flows for each processing step are described in the privacy policy and the processing register.
Files, the knowledge index and semantic search run on dedicated servers in Germany. Answer generation by Nex uses Mistral AI. Seven subprocessors, all named in the processing register under Art. 30: Netcup, Hetzner, Stripe Payments Europe and Mistral AI, plus Telnyx, Microsoft Corp. and Google LLC only when the respective feature is used. The figure of seven reflects the status as of 2 November 2026. The current version of the register is authoritative. In Nexoria's verified Mistral organisation, the use of new API interactions for model training is disabled (account checked on 26 September 2026). This setting applies to new interactions processed through that organisation. Separately connected AI accounts use their own account settings. Workspaces control who can access what.
Answers to typical questions from your review can be found in the Answers for your evaluation and under Security. A pilot with your own documents is described in our guide Introducing AI knowledge search.
Sources
- Regulation (EU) 2016/679, General Data Protection Regulation
- Regulation (EU) 2024/1689, Artificial Intelligence Act (EU AI Act)
- Regulation (EU) 2026/1744, Digital Omnibus on the EU AI Act
- Section 87 of the German Works Constitution Act (BetrVG)
- DSK, List of processing operations for which a DPIA must be carried out, version 1.1 of 17 October 2018
As of 15 November 2026.

