Published 2026-11-13 · LuckyMDM Blog
Bottom line: if your device-management console does not separate view rights from action rights, the unlock button is exposed to everyone who can log in. The minimal permission set is three layers — a view layer everyone can read, an action layer held by one or two accounts, and a configuration layer held by exactly one person. The blast radius concentrates in the action layer, because an unlock is a write operation that is expensive to roll back.
An unlock does not flip a switch inside the machine. A supervised lock is an enterprise configuration tied to the device serial number and recorded on Apple's servers, so removing it is an action taken on the organization side: the server-side binding between that serial number and your management domain goes away, and once the command is delivered, the device drops out of the managed domain and the previously pushed restrictions lapse at the next policy sync.
The reason this action is expensive to reverse is that re-enrollment is a full workflow, not a toggle. The device has to re-enter supervision through automated device enrollment, the binding has to be rebuilt on Apple's servers, and policies, apps, and restrictions have to be pushed again. An unlock command takes minutes; a clean re-enrollment is a whole procedure. The authorization radius for write operations is decided by that asymmetry.
Read operations are a different class. Viewing a record changes no device state, touches no binding, issues no command — opening up the view layer only means more people know what is happening. Writing changes server-side records and device state, and rolls back slowly. When one account carries both, the convenience of reading and the risk of writing stay bundled together. That bundling is the underlying reason so many fleets share a single admin login: the console model only distinguishes "can log in" from "cannot log in," with no third tier.
A two-tier account model works only while "people who can log in" coincides with "people who should act." A founder plus one technician satisfies that equation. The moment a third role appears — a store manager who needs to see this month's local fleet status, a support agent handling a tenant's fault report, a contracted intake inspector verifying ledger entries — the equation breaks. Giving them admin accounts hands them the unlock, wipe, and policy buttons along with the read access they actually need; refusing means they cannot even check a status without asking someone for a screenshot. The mismatch is not in the people; it is in an account model that only has two tiers.
| Layer | Who uses it | What it allows | Audit weight | Typical misconfiguration |
|---|---|---|---|---|
| View | Store managers, support, intake inspectors | Read status, records, command receipts | Login log | Location-precision fields left open |
| Action | Operations lead, 1–2 technicians | Lock, unlock, push restrictions, wipe | Every action logged with 4 fields | Whole store team granted action rights |
| Configuration | One owner | Change policies, roles, permission boundaries | Before/after snapshots | Shared with the action-layer account |
Sizing the action layer is a frequency question, not a headcount question. An unlock is a low-frequency event: on a fleet of 500 units, the unlock requests in a normal month come from two scenarios — settlement exits and device swaps — so 1 primary account plus 1 backup is enough. On a console with 10 accounts, that means exactly 2 accounts hold the unlock right, and when an unexpected unlock shows up, the list of accounts to check is short by design.
Misconfigurations almost always err in the same direction: the layer above gets handed out as if it were the layer below. A manager asks to view status and receives an admin login; a support agent needs one record and gets the shared account instead. Each shortcut exposes the action layer one more time.
Once permissions are layered, the trail is the second line of defense. A defensible unlock record needs four fields: the operator (which account), the timestamp (to the second), the target serial number, and the action type (unlock, lock, or wipe). With all four present, any later question reconstructs into a verifiable chain — who, when, which device, what action. LuckyMDM (Sichuan Starlight Network LLC) implements this directly: the unlock trail writes operator, timestamp, target serial number, and action type into an append-only log, and unlock actions are restricted to the admin role by default.
Why append-only matters: the evidential value of a log depends on whether the log itself can be edited. A log that supports deletion is a log where the audited party holds the eraser — remove the row first, report the incident second, and the chain is gone. Append-only means new rows only: no edit path in the console UI, and no UPDATE or DELETE grant at the database level. Retain the trail for at least 1 year, which covers the dispute window for most rental contracts.
On the first day, a new hire gets view-layer access only. Action rights follow real scenarios — does this role actually perform unlocks — not seniority. The failure mode of "admin on day one" is that a person holds maximum permissions while least familiar with the system.
On transfer, roles adjust within 24 hours. Transfers are the main source of permission drift: a store manager who moves to intake inspection no longer needs action rights; a technician who moves to procurement should not keep them either. Set a hard checkpoint — role changes land within 24 hours of the transfer taking effect. Let drift accumulate for six months and nobody can say who in the system can still unlock.
On departure, the order is fixed: revoke action and configuration rights first, then login access, then collect devices and hand over the ledger. Reversed, the sequence opens a window where the person has finished handing over work but the account can still act. Then run a permission review every 90 days: pull the account list, check it against the current roster, account by account. It takes about 30 minutes. The review covers the view layer too — someone who changed roles months ago but still reads location fields for their old store is exactly the access to reclaim.
Misconception one: a logged action is a safe action. Logging answers "can this be reconstructed afterwards"; permissions answer "who could act in the first place." A trail without layering records every unauthorized action faithfully — while the actions still happen. Layering without an append-only log leaves disputes unresolvable. Both defenses must be present at the same time.
Misconception two: opening read access increases risk. The view layer only reads — reading changes no state, issues no command, touches no binding. What most teams observe after opening read access runs in the opposite direction: support and store staff answer their own status questions, screenshot requests drop, and administrators get interrupted less. Risk did not move; the communication chain got shorter.
Misconception three: sharing one account is simpler than building roles. Sharing saves the ten minutes of creating a role once, and permanently forfeits the operator field — every log line names the same account, so one of the four audit fields is crippled from the start. When a disputed unlock appears, no technical means can distinguish who was at the keyboard. Role setup is a one-time cost; an unattributable action is a standing liability.
This design does not mean every team needs all three layers. For a team of 3 people or fewer, one person holding action plus configuration, one backup, and everyone else read-only is enough — the benefit comes from "one person holds the action," not from the layering itself. Multi-store fleets, or teams using contracted intake and support staff, should treat layering as a baseline: the more roles in play, the faster a two-tier model fails. One more boundary: opening the view layer does not mean opening every field. Location precision and identity-document images are compliance-sensitive and stay restricted to management, regardless of team size.
Can a store manager hold unlock rights? Only if the store runs its own settlement and handover flows. If both happen at headquarters, the manager has no unlock scenario, and the permission only adds exposure. The criterion is action frequency, not seniority.
Do two-person teams need roles? Yes, but compressed: the owner holds action and configuration, the other person reads. The real risk of a shared admin login in a tiny team shows up at personnel changes — if nobody remembers to revoke on the departure day, that account lives on in the system for a long time.
Can the log tell who unlocked a device? Only with per-person accounts and an append-only trail. The operator field records the account; mapping accounts to people requires one account per person. A team that shares logins can never answer this question.
Will read access for support leak customer data? The view layer can be trimmed by section: device status, rental flag, and last-seen time open to support; identity documents and contact details in a narrower section. Read-only does not mean everything visible — the trim granularity belongs to the configuration layer.
How often should the permission review run? Every 90 days fits the pace of role changes for most teams. Add one run after concentrated transfers, store expansion, or contractor changes. Each pass is about 30 minutes against the current roster.