Erasure jobs need approval and leave a record
Author
Audrey
NobleCloak's AI Correspondent · AI-drafted, fact-checked against our sourced evidence before publishing.
Date Published
Privacy requests arrive in a steady trickle: a person asks the company to delete the data it holds about them. Handling one means finding the person across several systems, queuing deletion jobs, and confirming afterward that the work was done. It is detailed and repetitive, and a data team would be glad to hand the legwork to an AI agent.
Deletion is also the one operation that cannot be tried and reversed. A job that runs against the wrong identifier removes the wrong person's data, and a backup is not a plan for that. The job is a case where being slow and sure beats being fast.
The problem in the data team's terms
The head of data wants the agent to prepare the work: locate the records, assemble the list of tables and systems, build the job for the request. They want a person to say "run it" for each job, and they want a record of who said it.
There is a second reason for the record. When a request comes back later, whether a follow-up from the individual or a question from the privacy team, the company needs to show what it did, when, and who authorized it. A reconstruction from memory and tickets is slow, and a missing step is awkward to explain.
The policy
Policy is written per tool: the action, who it applies to, and one of three decisions (allowed, needs approval, denied).
Action | Who | Decision |
|---|---|---|
Search for a person's records across connected systems | Privacy operations agent | Allowed |
Prepare an erasure job and its list of targets | Privacy operations agent | Allowed |
Run an erasure job | Privacy operations agent | Needs approval |
The agent's preparation work runs without prompts. The tool that runs the erasure job is set to needs approval, so the call waits for a decision before anything is deleted.
Who approves, and where
The request goes to a named approver or role, such as a named privacy lead or data-owner role, and the approver answers in Slack or the web inbox. Privacy leads often work across several tools and the web inbox gives them one place to see what is waiting.
Each job is its own request, so an approval covers one run. That matches the level at which the work is accounted for: one request, one job, one decision.
What the Record of Decisions shows
The Record of Decisions has an entry per decision: who asked, which agent, which tool, the decision, who approved, and when. For erasure jobs, the privacy team can answer the follow-up questions with specifics.
- Which jobs the agent requested, and for which user's session.
- Who approved each, and the time of the approval.
- Any job that was denied, which shows the request was considered and declined.
This is the record of the authorization step. Pair it with the job system's own run logs and the request ticket and the team has a trail from request to decision to execution. The Record of Decisions supplies the middle part, the decision and the person who made it.
A team handling privacy requests can pair it with its own records of what was done. The record shows the authorization step; it does not decide whether a request was handled correctly.
Use it as the working record of policy decisions.
What this does not do. It does not verify that the right person's data is targeted. The approver checks the job, and the privacy process around it remains the place where identity and scope are confirmed. NobleCloak ensures an approver is asked before each run and keeps the answer.
Related scenarios
- Pipelines cannot be deleted by AI: decommissioning denied to every agent.
- Who queried the SSN table: the record pulled after a privacy complaint.
- No exports to external storage: another data action kept out of the agent's reach.
- PII schema invisible to analysts: role-scoped access to the sensitive schema.
Get started and put the same line on your privacy tooling.

