A data subject access request lands in your inbox. Standard procedure — except this year, somewhere in your systems, AI agents have been reading, updating, and acting on that person's data. "Tell me what you hold about me" now includes "tell me what your agents did with it." Most DSAR runbooks haven't caught up.
The clock starts immediately
What makes DSARs operationally sharp isn't the principle — it's the statutory deadline. From the day the request arrives, you have a fixed window to respond, and "our agent activity is spread across four systems and an LLM provider's dashboard" is not an extension request the regulator accepts. The runbook question is brutally practical: given one name and one month, can you produce everything your automated systems did with that person's data, in a form you'd be willing to send them? For agent activity without a recorder, the honest answer is usually an engineering project. With one, it's an export.
The Article 15 answer
For access requests, the tape is the answer, packaged: a full-history subject export that runs asynchronously and downloads as a single artefact covering every record held for that subject. One detail says a lot about how it's built — any section that failed to read is reported in the export rather than silently omitted. A subject-access response with quiet gaps is a second complaint waiting to happen; this one states its own completeness, which is exactly the property you want to forward to a regulator with your response.
Finding one person in a sealed trail
Here's the puzzle unique to Wytness: the trail is pseudonymised, and we can't read it — so how do you find one subject's records? The answer is the elegant part. Because tokens are computed deterministically under your key, you can compute the subject's token yourself and match every record it appears in — without unsealing anybody else's data along the way.
Sit with that for a second, because it's the part that usually takes a re-read: the search that would be a privacy problem on most platforms is, here, a demonstration of the privacy architecture working. One person found, everyone else still sealed — and the platform that executed the search still doesn't know who anyone is. When a supervisory authority asks how you reconcile a searchable audit trail with data minimisation, this is the answer, and it's checkable rather than asserted.
The boundary between our job and yours
GDPR language matters, so we use it precisely, the same way our GDPR page does. Wytness is a processor for the audit data your agents generate — you are the controller, and you determine the purposes and means. Data-subject rights — access, erasure, portability — are operator obligations: we give you the tools to re-identify and respond, and we do not respond on your behalf. The Article 30 record of processing is likewise your DPO's document to assemble; the dashboard holds and exports the agent and tool inventory data it draws on. In plain terms: the subject writes to you, the answer goes out under your name, and our job is to make the assembling of that answer boring.
No GDPR pack — deliberately
Unlike SOC 2, the EU AI Act, and ISO 27001, there is no GDPR Evidence Pack, and that's a design decision rather than a gap. GDPR is evidenced by the machinery itself: the subject export answers Article 15 directly, the inventories sit behind Article 30, and for the security-and-minimisation articles the strongest evidence is the key flow — pseudonymisation at egress, customer-held keys, re-identification only on your side. When a supervisory authority asks how personal data is protected in your audit trail, the honest answer isn't a document. It's the architecture, and it demonstrates itself.
Before the first request arrives
The practical checklist: declare your PII fields so sealing actually covers them, keep your keys where your DSAR team can use them, and run one practice export before the real request arrives with a statutory clock attached. The rehearsal matters more than it sounds — the practice run is where you discover the field a team forgot to declare or the key that lives in the wrong vault, and each of those costs an afternoon to fix in rehearsal and a compliance incident to fix mid-request. The mechanics are in the pseudonymisation deep dive. Questions about your DSAR workflow? Ask us.