Table of contents
Is the handoff ready?
Use these checks before access expands or the first ticket begins.
Scope
Ready: The privileged paths and their blast radius are listed.
Needs work: Every admin tool is treated the same.
Evidence
Ready: Each session has a time, an account, commands, and a ticket.
Needs work: A verbal note that the work was approved.
Masking
Ready: Secrets and personal data are masked before storage.
Needs work: Raw tokens sit in a shared transcript.
Review
Ready: A named reviewer samples sessions on a published schedule.
Needs work: Sessions are opened only after an incident.
Decide which sessions need a recording
A privileged access session review starts with a list of access paths that can change production state. Include database consoles, cloud shells, admin panels, and deployment tools. Do not include read-only dashboards that cannot mutate a system. The list should be short enough to review at a weekly cadence.
Record the reason each path is privileged and the data it can reach. A session that can read customer records is different from one that can restart a queue. Rank the paths by blast radius rather than by tool popularity. The ranking tells the reviewer where to spend attention.
Capture the session without capturing secrets
Record the account, the start and end time, the command history, and the ticket reference. Mask tokens, passwords, and personal data before the record is stored. Keep a hash or a reference for values that must remain provable. The record should show the action without exposing the secret.
Use the platform's native session recording where it exists, and add a written note where it does not. Do not ask the contractor to record a personal screen video for every task. A manual recording is hard to retain and easy to edit. Prefer a system log that the client controls.
Pair every session with a ticket
Each session should point to an approved ticket with a stated scope. The ticket names the system, the change, the expected result, and the rollback step. A session without a ticket is a finding, not a routine exception. The reviewer should be able to match the log to the work without asking the engineer.
Record the approval before the session begins, not after it ends. A retroactive approval hides the moment when the risk was accepted. If an emergency forces a retroactive record, mark it clearly and explain why. The review should treat that case separately.
Review the session against the ticket
Compare the commands and the touched objects with the scope on the ticket. Look for reads of unrelated customer data, permission changes, or changes outside the named system. Note the difference rather than assuming it is harmless. A small unexplained read can matter more than a large planned change.
Ask the engineer to explain any step that the log cannot make clear. Do not rewrite the log to make the story tidy. Attach the explanation to the review record with the date and the reviewer. The explanation is evidence about the process, not a substitute for the log.
Sample and escalate findings
Review every session on the highest-risk paths and a sample of the rest. Set the sample size before the week starts so the choice is not driven by who is available. Record the sample method and the number reviewed. A sample that always covers the same engineer is not a sample.
Escalate a session that touched data or systems outside the ticket. Separate a misunderstanding from a control failure, and record which one applies. Give the engineer a chance to correct the record before the finding is closed. The outcome should change a permission, a ticket template, or a review step.
Report the review without exposing the log
Publish a short monthly summary with the number of sessions, the share reviewed, and the number of findings. Describe the findings by category rather than by person. Keep the raw log in the restricted store with its retention rule. The client should see the pattern without seeing customer data.
Close each finding with an owner and a date. Confirm that the permission change or the ticket change is live before the finding is marked done. Recheck one closed finding in the next review to test the follow-through. A review that never confirms its own fixes becomes a ritual.
Copy-ready brief
Put the operating agreement in writing.
Capture the account, the ticket, the start and end time, the commands, and the objects touched.
Replace tokens, passwords, and customer values with references before the record leaves the controlled store.
Review every session on a top-risk path and a declared sample of the remaining paths.
Close a finding only after the permission, ticket, or review change is live and confirmed.
Frequently asked questions
Which sessions should be recorded?
Every path that can change production state or read customer records, ranked by blast radius.
Should the contractor record a personal screen video?
No, prefer a client-controlled system log that is hard to edit and follows a retention rule.
What counts as a closed finding?
A live change to a permission, a ticket template, or a review step, confirmed by a second person.
Sources
- NIST SP 800-53 security and privacy controls
Access control and audit expectations.
- NIST SP 800-92 log management
Log retention and review practice.
- CISA Secure by Design
Accountability and least privilege.
- Google SRE Workbook
Review and operational evidence.
