Every engineer eventually inherits a system where nobody can explain who may do what. The symptoms are familiar: a long list of administrators, requests approved in seconds, an emergency login nobody has tested, and an audit coming up. This article treats that situation as a diagnostic exercise. We list the symptoms, trace each to a cause using the Ops Atlas description of AccessOps, and pair every cause with a fix. It is aimed at learners and early-career engineers who want a repeatable method rather than a list of tools.
Symptom 1: Too many people can change production
Likely cause: permissions were handed out for a one-off project and never taken back, so the list only ever grows. The Atlas calls the healthy alternative an eligible-not-active model, where people qualify for powerful roles but hold them only when a task requires it.
Diagnosis step: export the list of holders for the top two tiers of systems and count them. In the Atlas worked example, standing accounts at those tiers fell from thirty-four to nine, then two emergency uses left seven to explain. Your baseline matters more than the example.
Fix: convert most holders to eligible, require a reason on activation, and keep a short list of permanent emergency accounts. The thinking behind this method is close to what we teach in our guide on sharpening your analytical habits when solving DevOps problems: measure first, change one variable, measure again.
Symptom 2: Approvals are instant and meaningless
Likely cause: one manager or one rule approves everything. The Atlas lists blanket manager approval as an anti-pattern because it breeds generic justifications such as “needed for work”, which tell reviewers nothing.
Diagnosis step: sample ten recent requests and read only the reason field. If you cannot tell what happened, the approval step is not a control.
Fix: give each grant record a fixed set of five columns, with the asker, the resource, the task behind it, the expiry and the sign-off. Require a task-shaped reason. Send top-tier requests to two approvers, and let the person on call approve urgent ones within a short limit with a next-day review.
Symptom 3: Access is revoked, but things still work
Likely cause: you are assuming expiry terminates sessions. The Atlas draws a sharp line between holding a grant and running a session. In its lab, a remote shell stayed connected for about twenty seconds after the certificate expired, since the tool checks only at login. Cloud role sessions can persist until their natural expiry unless revoked, and a major cloud provider’s documentation describes permission propagation of about two minutes, sometimes seven or more, with group changes taking much longer. Roles that can edit permissions may also leave access that survives the original expiry.
Diagnosis step: issue yourself a short grant in a test environment, keep a session open, and watch what happens when the grant ends.
Fix: shorten session lifetimes, add explicit revocation or session-termination steps and restrict who can edit permissions in the first place.
Symptom 4: Emergency login exists, but nobody trusts it
Likely cause: it has never been exercised. The Atlas, citing one major vendor, asks for at least two cloud-only emergency accounts with phishing-resistant sign-in, an alarm on every login and a working-order check roughly once a quarter. These accounts are permanently active by design, which is why they need alarms and drills.
Diagnosis step: try to sign in with the account from a clean device and time it. Count how many people know where the credentials live.
Fix: schedule a quarterly drill, store credentials in two places, review each real use within one business day and write the rule for when the account may be touched.
Symptom 5: We have a vault, yet standing rights remain
Likely cause: the tool was installed but the practice was not. The Atlas is blunt that owning such a product proves little about how access is actually run. Another trap it mentions is infrastructure code that authoritatively overwrites bindings, silently erasing settings the vault relied on.
Diagnosis step: compare the number of standing privileged accounts before and after the tool arrived.
Fix: define target outcomes, like the share of access issued just in time and the percentage of privileged sessions recorded, and review them monthly. Also check how your templates manage role bindings, so one tool does not undo another.
Symptom 6: Reviews find problems every year, but they return
Likely cause: review is carrying the whole load. The Atlas notes the federal maturity guidance sees permanent access plus review as the Traditional stage, not the target. Reviews are still useful, and a controls list it cites asks for at least one a year, but they cannot be the only defence.
Fix: use each review to identify rights that should become eligible-on-request. Flag dormant accounts early; one list suggests forty-five days, while another vendor’s review tooling recommends denying after thirty days without sign-in.
Reading the symptoms together
The six symptoms rarely appear alone. A team with too many administrators usually has weak reasons on requests, because nobody ever had to justify the original grant. A team with an untested emergency account tends to lean on it for convenience, since the normal path is slow. Look for the shared root: access was treated as a one-time setup task, not as an operation with a rhythm of requests, expiry, review and learning.
That shift in mindset is the practical meaning of the AccessOps label. Choose one symptom this month, apply the diagnosis step, and write down what you found. Small, documented experiments teach more than a big redesign launched on a hunch.
Putting the diagnosis in order
A sensible order for a newcomer is to count, then reason, then expire, then drill, then measure. That sequence matches the Atlas thirty-sixty-ninety outline, which starts with inventory and an emergency drill, moves to a pilot of eligibility with reasons and on-call auto-approval, and ends with dual approval for the highest tier, recording and daily reconciliation.
For wider career context, our article on DevOps skills and strategy for IT transformation explains where operational habits like these fit in a learning roadmap. The entry in the Atlas covering AccessOps is a handy reference to keep open while you work through the checks above.
A note on evidence
Treat every number in this article as an example. Your estate, your regulators and your tools will set different limits. The point of the diagnosis is to replace opinions with observations: count the holders, read the reasons, test the expiry, drill the emergency path, and decide from what you see.

