GDPR Article 30 records of processing for AI agents

Your GDPR Article 30 record of processing activities must cover AI agents, but not as a single line. An agent is a means, not a purpose. Each purpose it serves needs an entry that lists the categories of people and data, the recipients, including the model provider, any transfers, retention for logs and memory, and the security measures.

What does Article 30 require?

Article 30 requires controllers, and processors in their own form, to keep a written record of processing activities under their responsibility. In substance, a controller's record must contain:

• the name and contact details of the controller and, where applicable, joint controller, representative and data protection officer;
• the purposes of the processing;
• a description of the categories of data subjects and of personal data;
• the categories of recipients to whom personal data has been or will be disclosed, including recipients in third countries;
• transfers to third countries or international organisations and, for some, the safeguards;
• where possible, the envisaged time limits for erasure of the different categories of data;
• where possible, a general description of the technical and organisational security measures.

The record must be in writing, including electronic form, and made available to the supervisory authority on request. There is a limited exemption for small organisations, but it does not apply where processing is not occasional, is likely to result in risk, or includes special category data. Agents processing customer data will rarely fall within it.

Article 30 is the accountability principle made concrete; see GDPR Article 5 principles applied to AI agents.

Who must keep the record?

Every controller and processor within scope. A firm deploying agents on its own customers' data is a controller and keeps the controller record. A platform that runs agents for clients is usually a processor and keeps a processor record listing each controller it acts for, the categories of processing, transfers and security measures.

Why do agents need special care in the record?

A traditional system usually maps to one or two processing activities. An agent often serves several purposes, reads many sources, sends data to a model provider, calls third-party tools and keeps logs and sometimes memory. A single entry called "AI agent" fails the record's purpose, because no one reading it can tell what data is processed, why, or where it goes.

Five points are specific to agents:

1. Purposes multiply. A servicing agent might handle account queries, collections reminders and complaint intake. Each is a purpose with its own lawful basis, and each needs its own entry or its own line within an existing entry.
2. Data categories expand quietly. An agent with access to a full customer record processes every category in it whenever it retrieves the record, even if the task needs one field. List what it can access, and work towards making that match what it needs.
3. Recipients include the model provider. Prompts containing personal data are disclosed to the provider. Tool providers the agent calls, such as an address lookup or payment service, are recipients too.
4. Transfers follow the model. If inference runs outside the EU, that is a transfer to be listed with its safeguard.
5. Retention has new categories. Action logs, conversation history, retrieved document caches and agent memory each need a retention period. Memory is often forgotten; our answer to whether GDPR applies to what an AI agent remembers is that it does.

How do you add agents to the record?

The last row is where most records fail. Agents change faster than annual record reviews. Linking the record to the agent's change control keeps them aligned.

The lawful basis column depends on the work set out in GDPR Article 6 lawful basis for AI agents processing customer data.

What will a supervisor or auditor ask?

• Show us where your agents appear in the record of processing.
• For this agent, what purposes does it serve, and what data can it access for each?
• Who receives personal data from your agents, and where is it processed?
• How long do you keep agent logs and memory?
• When did the record last change, and what triggered it?

What does Article 30 not cover?

It does not require a record per agent; it requires a record per processing activity. It does not replace the DPIA, which assesses risk in depth. It does not set retention periods; it records the ones you set. And it does not cover the EU AI Act's register of AI systems, which serves a different purpose, though maintaining both from the same source of truth is sensible. For the full map, see AI agent regulation, mapped to controls.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Where Pryme fits

In Pryme, each agent's sources, connections and role scope are defined per capability, so the data an agent can reach is a configuration you can export, not a guess. Every fact arrives with its provenance class. That makes the categories and recipients columns of the record something you can generate rather than estimate.