How to Add AI Tools to Your Record of Processing

Art. 30 GDPR requires most companies to keep a record of processing activities, and every AI tool that touches personal data belongs in it. Dr. Datenschutz published a guide to the record in November 2026, which gives you a reason to check how your AI tools are documented.

Sandra Leistner4 min read

Illustration: A card catalogue with an open drawer holding a gear among blank cards, beside a closed ledger book held upright by a bookend.

Why the record comes before any AI rollout

The record of processing activities (in German, Verzeichnis von Verarbeitungstätigkeiten, or VVT) lists which personal data a company processes, for which purposes and with which safeguards. Dr. Datenschutz explains that it serves internal transparency and that a supervisory authority can ask to see it. The authority expects a record that already exists and is kept up to date, so there is no grace period to build it afterwards.

AI tools make this concrete. If the record does not mention them, the company cannot say what happens to that data. It also cannot answer a data subject's request quickly.

Does the exemption for firms under 250 staff apply?

Art. 30(5) GDPR contains an exemption for organisations with fewer than 250 employees, but it has conditions. The processing must be only occasional, and it must not involve special categories of data under Art. 9(1), such as health data, or data on criminal convictions under Art. 10. Dr. Datenschutz points out that regular processing is the norm today, because even small firms keep personnel files electronically and run their correspondence by email.

An AI tool used every working day is regular processing by any reasonable reading. Where staff paste health information or similar data into it, the exemption falls away regardless of headcount. Most Mittelstand companies, the German small and mid-sized businesses, should therefore plan on keeping a record.

What an entry for an AI tool should contain

Art. 30(1) GDPR lists the mandatory content, and an AI entry follows the same structure. The points below are applied to AI use:

  • Controller, data protection officer and, where relevant, representative with contact details.
  • Purpose, for example drafting replies, transcribing meetings or searching internal documents.
  • Categories of data subjects and data, such as customers, employees or applicants, and the data types that realistically enter prompts or documents.
  • Categories of recipients, which includes the AI provider and any sub-processors.
  • Transfers to third countries and the safeguards used, where they occur.
  • Planned deletion periods, covering prompts, outputs, logs and recordings where possible.
  • A general description of technical and organisational measures, or a reference to your existing document on them.

Supervisory authorities also recommend noting the legal basis and extra safeguards for each activity. For AI this helps, because it shows that someone checked whether the use is lawful before it started. Existing concepts need not be copied, and an attachment or a reference is enough.

One processing activity or several entries?

A processing activity groups operations that serve the same purpose, so the record does not need one line per processing step. The guide notes that very small units may combine whole departments, while larger firms often list processes separately. It also suggests separate entries for systems such as time recording, CRM or HR software.

For AI, the purpose is the useful dividing line. A meeting transcription tool and an assistant that answers questions from the document archive serve different purposes, involve different people and carry different risks. Give each its own entry. If HR and sales both use the same assistant, check whether the data and the purposes differ enough to justify two entries.

Step by step: adding an AI tool

  1. List every AI tool in use, including those staff adopted on their own, such as browser chatbots or built-in assistants.
  2. Ask each department what data goes in, who sees the output and where results are stored.
  3. Check whether special categories of data can appear and note this in the entry.
  4. Identify the provider, its sub-processors and the hosting location, and record any third-country transfer with its safeguards.
  5. Set deletion periods for prompts, outputs and logs, and link your existing measures document.
  6. Name an owner in the department, and review the entry whenever the tool, model or purpose changes.

The data protection officer advises on this, but responsibility stays with management, and the detail comes from the departments that know the workflows. Rules on professional secrecy, such as section 203 StGB for lawyers and tax advisers, and the AI Act may add duties that sit outside the record itself.

Using AI with company documents in a way you can document

It reads only, respects the existing permissions, and cites the document, version and page for each answer.

The same software runs in two modes: On Demand, hosted in Germany, or on-premise as Private Spark in your own network. The model catalogue includes local models, models hosted in Germany and external EU models with zero data retention, all listed on the model catalogue page. Agents prepare tasks, and anything that writes to a system waits for approval. The page about GDPR-compliant AI covers the data protection side in more detail.

Sources

More articles

All articles

How safe are AI API routes via resellers and routers?

Security firm Dragos sent about two million requests with fake credentials to AI providers and found compromised services and resellers reading customer prompts. Companies that use AI through third parties need to know who sits on the path of their data.

Data protection4 min

Run Text Generation Inference with Docker on Your Server

Text Generation Inference (TGI) serves open language models from a single Docker container on your own hardware. This tutorial starts it, sends requests over HTTP and keeps prompts inside your network.

Tutorial4 min

Cloud outage liability: who answers when Microsoft fails

On 12 November 2026 a software fault took Exchange Online and Outlook offline for about 36 hours. A legal analysis published shortly afterwards argues that responsibility for such outages often lies with the operator rather than the provider.

Regulation4 min

Test it with your own documents.

Free for 3 days, then from €24.90 per user per month. Cancellable monthly.

Free for 3 days, then from €24.90 per user per month.Free for 3 days
Try free for 3 days