A Flight Academy’s Ninety Days of AccessOps: A Team Diary

A Flight Academy’s Ninety Days of AccessOps: A Team Diary

The team below does not exist. It is an invented, generic flight academy with a scheduling system, a records database, a payments page and four people who look after all three. Their story is a way to walk through the order in which a small organisation might apply AccessOps, with each decision set beside what published guidance says. Treat every figure inside the story as illustrative, since it was invented to make the sequence easy to follow. The facts about standards and incidents come from the Ops Atlas entry, its page on AccessOps for privileged access.

Day 1: A discovery nobody liked

The team lead asks a simple question in the weekly meeting: who can export every student record? Four answers come back, then a fifth from a contractor’s email thread, then a long silence. In the story, eleven accounts turn out to qualify. Three belong to people who left the academy.

The Atlas treats this stage as the bottom of its maturity ladder, where nobody has an inventory. Its suggested first thirty days are blunt: inventory the two most powerful tiers and run an emergency-account drill. The team writes the eleven names on a whiteboard and, for each, a one-line answer to five questions: who holds it, what it opens, why, until when, and who approved it. Seven rows cannot answer the last two.

Day 8: The shell that would not die

The team’s systems engineer decides to test the end-date idea on a staging server. She grants a colleague a short-lived certificate to log in, lets the colleague connect, and waits for the expiry. The session carries on. A few seconds later it is still open.

The story echoes a lab the Atlas describes in which a remote shell outlasted its certificate by about twenty seconds, because the software checks at login only. The team writes a note: expiry stops new logins; it does not close existing ones. They add a step to their procedure, which is to end live sessions when the window closes. This is the first lesson the entry hammers home, that a grant is not a session.

Day 15: Windows with numbers on them

Next, tiers. The registrar’s database is the most sensitive, the scheduling system sits a rung below, and the marketing site is at the bottom. The Atlas suggests ceilings that step from one-to-two hours at the summit down through two-to-four and four-to-eight, ending in standing access with a yearly check. The team picks the two-hour end for the registrar’s data and four hours for scheduling. Their reasoning is evidence-led but modest: they will measure how long jobs really take and tighten later.

That measurement matters. The entry’s worked example shows a month in which 400 four-hour grants produced 96,000 granted minutes, while recorded use amounted to 24,000. In the diary, the team’s own first month is more embarrassing: they have so little recording that they cannot calculate utilisation at all.

Day 30: Emergency drill

The drill goes badly, which is what drills are for. The emergency account’s password has been changed by someone who thought it was a stray login. The team rebuilds the pair of emergency logins in the cloud directory, switches on phishing-resistant authentication, attaches an alarm to any use, and books a recurring validation every three months, the longest gap the entry attributes to one major vendor’s guidance. A rule is added that anyone who uses one must file a note within a business day.

A new colleague asks whether this defeats the purpose of removing permanent rights. The lead shows him the Atlas’s explanation: zero standing privileges does not mean zero administrators, and the emergency accounts are meant to stay enabled for good.

Day 45: The reason field

The team introduces eligibility for the scheduling system’s administrators. Staff no longer hold admin rights; they ask, type a reason and switch them on for a few hours. In the first week the reasons read “need access” and “updates”. The Atlas warns that blanket approvals make justifications generic. So the team changes the form to ask for the student cohort or the booking reference involved. Reasons improve within days.

In an aside, the lead rereads two earlier posts on this very blog about disciplined operations. One covered privacy habits for high-profile charter passengers, another listed typical safety errors that charter customers make. Neither concerns computers, but both argue for the same thing: write down the sensitive steps, and do not rely on goodwill.

Day 60: The approver who was nobody

Someone sets up an approval rule with a list of approvers and then removes the last name during a clean-up. The next request cannot be approved by anyone. The team is lucky to catch it in a test. The Atlas records exactly this kind of mistake, an empty approver list locking a tenant out, among its anti-patterns, along with the reverse risk of infrastructure code that overwrites permission settings on every run. The team adds a check to the change process and makes sure the emergency account is the way back in.

Day 75: Reviewing the reviewers

A suggestion comes up: run a big quarterly review and call it done. The lead is wary. The federal maturity model that the Atlas follows treats standing access plus periodic review as the Traditional stage, not the goal. The CIS Controls ask for reviews at least annually at their strictest group, but the Atlas presents them as a backstop. The team keeps a lighter monthly look at dormant accounts instead. One benchmark flags idle accounts at forty-five days, and a vendor review tool suggests denying after thirty. They pick thirty.

Day 90: Taking stock

Eleven top-tier accounts are now three eligible administrators and two emergency accounts. Every grant has five fields filled in. Contractors get windows that end by themselves. Recording exists for registrar exports. Nothing dramatic has happened, and nothing has been prevented that anyone can point to, which is how this work often feels.

What the story shows

  • Order matters: inventory, time limits, reasons, then approvals and recording.
  • Test the assumption behind every control. Expiry and session end are separate events.
  • Emergency routes need maintenance as much as normal ones.
  • Reviews support the work; they are not the work.

Real teams will see different numbers. The sequence, though, is borrowed from the public record, and any small academy could start the same way tomorrow morning.