Walk into any engineering channel on the day a privileged-access project is announced and you will hear confident slogans: “We’re going zero-standing, so nobody is an admin.” “Just-in-time fixes stolen credentials.” “We bought the vault, so we’re covered.” Some of these are half true, which makes them more dangerous than plain falsehoods. Here are six AccessOps myths, checked against the Ops Atlas and the incident cases it cites, with a plain verdict on each.
Myth 1: Zero standing privileges means zero administrators
Verdict: false. The goal is to avoid permanent, always-on rights for ordinary administrative work, not to abolish the people who administer. The Atlas stresses that emergency accounts remain active permanently by design, since they exist for the moment the approval machinery itself is down. If your design ends with literally nobody able to act during an outage of the identity system, you have built a lockout, not a security posture.
What to do instead: keep two or more emergency accounts that depend on nothing else, protect them with phishing-resistant sign-in, alarm every use and test them at least every ninety days, the interval one major cloud vendor asks for.
Myth 2: Just-in-time access stops stolen sessions
Read the Ops Atlas treatment of AccessOps and the position is blunt: shorter windows shrink exposure, but a session that has already been established keeps its power. In its lab, a remote shell kept running about twenty seconds after the certificate behind it expired, because the tool inspects certificates only at login. Cloud role sessions can also stay valid until their own expiry unless someone revokes them, and permission changes can take minutes, in some documented cases seven or more, to propagate.
Verdict: partly true. JIT reduces how long a stolen identity is valuable. It does nothing about a thief who is already inside a session. Pair it with short session lifetimes, explicit revocation and monitoring.
Myth 3: We bought a vault, so access is managed
Verdict: false. A privileged-access product is an instrument, not a practice. The Atlas lists this under anti-patterns: tools sit half-configured while accounts keep standing rights, approvals become formalities, and recordings pile up unwatched. A useful test is to count standing privileged accounts before and after the purchase. If the number has not dropped, the vault is a very expensive safe sitting in an open building.
Project scoping matters here too. Our guide on engineering blueprints for web project requirements makes the same argument for software generally: write down the outcomes you need first, then judge whether a tool delivers them.
Myth 4: An annual access review is our main control
Verdict: false, and the federal maturity model agrees. According to the Atlas, the American zero-trust maturity model files an arrangement of lasting rights topped with periodic review under its opening stage, named Traditional. Reviews are valuable as a safety net, and a yearly cadence is a common minimum. As the main defence they are slow, subject to rubber-stamping and blind to what happened between checks.
Better: use reviews to find grants that should not exist, then convert those to eligible-on-request. Track how many standing Tier 0 and Tier 1 accounts you hold and watch the line trend down.
Myth 5: Break-glass is fine as a shortcut when the process is slow
Verdict: false. The Atlas quotes a cloud provider whose own guidance cautions that this kind of login may amount to a backdoor, a warning aimed at precisely this habit. When normal requests take too long, fix the request path: faster approval for on-call staff, pre-approved rules for routine tasks, sensible defaults. If the emergency account shows up in every weekly report, the alarm is telling you the normal path is broken.
Record the lag between use and review as a metric. The Atlas suggests reviewing each emergency use within one business day.
Myth 6: An automated approver can safely wave through urgent requests
Verdict: dangerous when unexamined. The Atlas treats any automated approver as a sensitive system in its own right, equal in weight to the access it protects. It gives a concrete example: one open-source access platform’s paging integration only looks up whether the requester appears on the duty roster, and never confirms that an incident is actually open. So any rostered person could, in principle, ask for elevation at any hour of the week. The entry also says no documented safe design yet exists for letting a language model approve top-tier requests; such models may summarise a request for a human, and that is about all.
If you want automatic approval for urgent work, define exactly what the rule proves, narrow the scope to the tier below the crown jewels, and send every automatic approval to a reviewer afterwards.
How to test a belief before it becomes policy
Myths survive because nobody checks them. A cheap routine for any team is to turn each belief into a falsifiable statement and look at the data once a month. “Nobody holds standing Tier 0 rights” can be checked with one query. “Emergency accounts are used rarely” can be checked against the alarm log. “Requests are justified” can be checked by sampling ten recent reasons and asking whether a stranger could understand them.
The Atlas worked example is a good model of this style. It describes a team whose standing Tier 0 and Tier 1 accounts dropped from thirty-four to nine, with two emergency uses leaving seven to explain, and it shows how an approved-but-unused figure of thirty out of one hundred twenty, or a quarter, exposed requests that people had filed out of habit. Numbers like these cut through opinion. They also keep the project honest when someone cheerfully announces that the problem has been solved.
One more habit from the myth list: never let a single incident story become the whole argument. Cases cited in the Atlas include a 2020 social-media breach in which more than a thousand employees could reach account tools, and a 2023 build-platform incident where a stolen, already-authenticated session inherited broad token rights. Both illustrate the cost of wide standing reach. Neither proves that any single technique would have prevented the loss. Use cases to ask better questions, not to sell a product.
One thread through all six AccessOps myths
Each belief on the list swaps a mechanism for an outcome. “We have the vault” is a mechanism. “Nobody holds standing rights unexplained” is an outcome. Before adopting any control, ask which outcome it produces and how you will measure it. The Atlas offers a short list worth stealing: standing privileged count, share of access granted just in time, median and ninetieth-percentile time to access, grant utilisation, review lag for emergency use, and recording coverage.
Before building anything custom to handle all this, read our piece on the full list of questions to settle ahead of any bespoke web application. Half the value of a good access project comes from asking boring questions early: who needs this, for how long, and who will notice if it goes wrong?

