Early in training, a student is sent solo for the first time. The instructor has to decide how much trust that means. One school gives a blanket endorsement: you may fly the local area in this aircraft until further notice. Another clears each flight individually: today, this route, this weather, back by noon. Both schools produce safe pilots on good days. Which one holds up on a bad day is the interesting question, and it is the same one the Ops Atlas asks about computer access in its AccessOps entry.
The two models in plain terms
The Atlas, on its page about AccessOps and privileged access, puts the contrast like this. In the first model, people hold powerful rights permanently, and the organisation checks periodically that this is still appropriate. In the second, people hold only the ability to request rights, and each activation is temporary and reasoned. It calls the second an eligible-not-active arrangement and sets zero standing privileges as the target for the most sensitive work. The first is the season-long endorsement; the second is the per-flight sign-off.
Round one: How much can go wrong while nobody is looking
The blanket endorsement is cheap for everyone on a normal day. Nothing to ask, nothing to approve. The price shows up when circumstances change. A student who was ready in March may not be ready after a long break, yet the endorsement is still on file. In the computing world, that is the account that kept its power after the person changed roles. The American government’s zero-trust maturity scale, which the Atlas lines up against, files that arrangement, always-on rights with a periodic look, under its Traditional stage, the lowest of the levels.
The per-flight sign-off keeps the exposure window short. If something goes wrong on Tuesday afternoon, the authority has already lapsed by Tuesday night. The Atlas suggests starting windows from one or two hours at the top tier up to four to eight at the third, with the lowest tier remaining standing and reviewed annually. The comparison favours the second model on exposure, though not absolutely: a stolen live session is still usable until it ends, and the entry says plainly that just-in-time access is not attacker-proof.
Round two: Friction and the human cost
Here the endorsement wins. Anyone who has watched a student wait for a busy instructor to sign a dispatch sheet while the weather window closes understands the problem. Per-flight sign-off adds delay, and delay changes behaviour: people hurry, copy credentials or find a way around the process. The Atlas’s own metrics include the median and the slow-tail time to get access for this reason. A fast median hides a slow tail that is, in practice, where workarounds are born.
It also cites an example in which activation of a secure route slowed an outage response at a large social platform in 2021, and the organisation judged the delay acceptable. The point is not that friction is always good; it is that someone should decide on purpose which jobs deserve it. For students in our own world, a piece on staying composed when training pressure builds makes a similar argument: preparation removes the panic that bad shortcuts feed on.
Round three: Cost and effort
The blanket model costs little to set up. The per-flight model needs tooling, rules for approvers, an on-call process and a way to measure use. It also needs maintenance. The Atlas cautions that having bought a vaulting product proves nothing about how access is really handled, and it warns about approver lists that end up empty and lock the whole system. If you are weighing cost carefully, our practical article on stretching a flight-training budget sensibly is a good model for the mindset: spend where the risk is highest first.
A fair accounting also includes waste in the temporary model. The entry’s example month finds that four-hour windows were granted 400 times and used for a quarter of that time, so the default was four times the real work. Short grants are not automatically right-sized.
Round four: Evidence when something goes wrong
A season-long endorsement tells you little after an incident beyond the fact that somebody had permission. Per-flight records tell you who flew, when, why and on whose say-so. The Atlas’s five-field grant record (holder, scope, reason, end time, approver) is the digital version of a dispatch sheet. Recording sessions adds a flight-data log on top, with a note that capturing everything and sampling nothing is just another way to be unprepared.
A small illustration of the middle path
Imagine a school that runs a booking system, a student records database and a payments page. The booking system is used all day by front-desk staff for scheduling, so a standing role there, reviewed once a year, is reasonable. The student records database holds identity documents and training histories, so the school treats it as a higher tier: staff can request access for a couple of hours, give a reason that names the student or task, and a rule approves it if they are on the roster. The payments page gets the strictest handling, with a second person approving and sessions recorded. The same people, three different kinds of trust.
So which should a school choose?
Neither purely. Most well-run schools mix the two. Routine, low-risk activity gets a standing endorsement with an annual look; the highest-risk activity gets per-flight sign-off. The Atlas expresses the same blend as tiers, and its maturity path climbs from improvisation, via a counted inventory and expiring grants, to on-request elevation as the default and, finally, rights scoped to a single task.
If you are starting from the endorsement side, do not leap straight to a full redesign. Take its thirty-day step: list who holds the most powerful rights and rehearse the emergency account. Then add end dates, then reasons, and finally approvals with recording for the very top.
One thing both models need
An emergency procedure that skips both. A flight school has a plan for the day the dispatch desk is dark. A computer team needs break-glass accounts that stay enabled by design, are tested at least every ninety days per the vendor guidance the entry cites, and raise an alarm when used. Without it, the strict model can lock out the very people it is meant to serve.
Pick the model that matches the risk, write the reason down, and revisit the choice when the weather changes.

