Data Subject Request Preparation Agent · Use case
How can we use an AI agent to prepare the reply to a data subject request on our own infrastructure?
A data subject request preparation agent takes one person's request for their data, finds the records in scope in the systems you have connected, lists what may need to be withheld and why, and drafts the reply. The privacy officer decides what is released and sends the reply. The agent confirms no one's identity, releases nothing, deletes nothing and decides no exemption. It runs on your own infrastructure, so the person's records are not sent to a public AI service.
One request.
Records in a dozen systems, and a date set by law.
- ✗ 01A request for a person's data means searching every system that may hold it, and the reply is due on a date the law sets.
- ✗ 02The records found name other people and hold confidential material, so each one is read before it is released.
Where it stops for a person
Always. The privacy officer decides what is released and sends the reply.Always. The privacy officer decides what is released and sends the reply. The agent proposes reasons to withhold, it does not apply them: whether an exemption applies is the privacy officer's decision, with a lawyer where one is needed. It lists the systems it searched, so a system it could not reach is visible, and a search that finds nothing is reported as that and not as "no data held". Confirming the requester's identity is a person's step.
It gathers and drafts.
The privacy officer decides and sends.
Step · 01
The privacy team uploads the request, or opens it from the ticket, once a person has confirmed who the requester is.
Step · 02
The agent reads what is asked: whose data, which kinds, which dates.
Step · 03
It searches the systems you have connected, read only, and lists the records in scope, with where each was found.
Step · 04
It lists the records that may need to be withheld or redacted, with the reason: another person is named, a lawyer's advice appears, a confidential matter is described.
Step · 05
It drafts the reply in your form, and the privacy officer decides what is released, makes the redactions and sends it.
Reads from · connectors
Your own APIs
The request, as a ticket or an uploaded file; the records that may hold the person's data, read through your own APIs with the operations that change something left out, or as a knowledge source; and your procedure and reply forms, indexed as a knowledge source. It searches as the privacy team member using it, and no further.
Acts through · MCP servers
Nothing. It gathers, lists and drafts. It sends no reply, releases no record, redacts and deletes nothing, and changes no system.
What you set up
- List the systems that hold personal data, and import the description of each one's API, keeping only the operations that read.
- Index your procedure, your reply forms and your guidance on withholding as a knowledge source.
- Build the agent in Katonic Studio and test it on past requests, including ones where records were withheld.
- Publish it to the privacy team only.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service. The request and every record found stay inside the company.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does the agent send the reply to the person?
No. It drafts the reply. The privacy officer decides what is released and sends it.
Does it decide what to withhold?
No. It lists records that may need to be withheld, with the reason. The privacy officer decides, with a lawyer where one is needed.
Where are the person's records processed?
On your own infrastructure, with a model you serve. Nothing is sent to a public AI service.