GDPR data subject access requests and AI agent logs

AI agent logs usually fall within a GDPR subject access request. If a record of what an agent read, wrote or decided contains personal data about the requester, they are entitled to a copy, along with information on purposes, recipients, retention and any automated decision-making. Finding and redacting those records is far easier if logs are designed for it.

What does the right of access require?

GDPR gives people the right to obtain from a controller confirmation of whether their personal data is being processed and, if so, access to it and certain information. In substance, the controller must provide:

• a copy of the personal data being processed;
• the purposes of processing;
• the categories of personal data;
• the recipients or categories of recipients, including any outside the EU;
• the retention period or the criteria for it;
• the person's other rights, including rectification, erasure, restriction and objection, and the right to complain to a supervisory authority;
• the source of the data, where not collected from the person;
• the existence of automated decision-making, including profiling, and in those cases meaningful information about the logic involved and its significance and envisaged consequences.

The response must be given without undue delay and within a fixed statutory period, extendable in some circumstances. The copy must not adversely affect the rights and freedoms of others, which is the basis for redacting third-party data.

Who must respond?

The controller. Processors, including agent platform providers, must help the controller respond. Agents can help too, by searching and assembling material; the playbook on how to handle data subject access requests with an AI agent covers that use. Whether an agent may run the whole process is a separate question, answered in whether an AI agent can handle a data subject access request.

What in an agent's records is personal data about the requester?

More than most firms expect. An agent's records may include:

• Conversation transcripts with the requester.
• Action records naming them: messages sent, records changed, payments prepared, decisions held for approval.
• Retrieved content: copies or excerpts of documents about them that the agent read.
• Agent-generated content: summaries, notes and inferences about them, such as a vulnerability flag or a risk assessment.
• Memory: facts the agent stored about them across sessions.
• Prompts sent to the model provider containing their data, if retained.

Inferences are the item most often missed. If an agent concluded that a customer is in financial difficulty, that conclusion is personal data about them, and the request reaches it.

The information duties also bite. If an agent made solely automated decisions with significant effects about the requester, the response must give meaningful information about the logic. If the agent's records went to a model provider outside the EU, that is a recipient and a transfer to disclose.

How do you retrieve and redact agent records?

The first row decides everything else. If agent records are tagged at the point of creation with the people they concern, a request becomes a search. If they are not, it becomes an excavation, and deadlines will be missed.

Tiered logs help. A pseudonymised event record and a short-lived content tier, as described for retention purposes, also make access requests manageable: the event record shows what happened, and the content tier holds the personal detail for a shorter period.

What will a supervisor or auditor ask?

• Do your access responses include agent records? Show us one.
• How do you find all records an agent created about a person, including memory?
• How do you handle third-party data in agent transcripts?
• How do you explain automated decisions in a response?
• How long does retrieval take?

What does the right of access not require?

It does not require disclosure of trade secrets or the model itself, though these do not justify refusing to give the personal data. It does not require raw logs if the data can be provided in a clear, complete form. It does not extend to other people's data. And it does not require you to keep data in order to answer future requests; storage limitation still applies. The related right to have data erased is covered in GDPR erasure requests and AI agent memory, and the general handling of personal data in AI agent logs applies here too. 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

Pryme files every verdict in an append-only record linked to the case and the people it concerns. Facts carry their provenance class, so a response can show the requester which data came from your systems of record and which came from them or from outside.