Kill Standing Privilege
Updated
IntermediateWhat you'll learn
~15 min- Convert permanent privileged assignments to just-in-time eligible roles
- Design the activation gate, the review cadence, and the tiered-admin boundary
- Operationalize assume-breach: privilege that expires limits what a stolen session can do
Assume breach, applied to your own admins
Tenet 5 said assume nothing is clean — including, uncomfortably, your administrators’ sessions. The question that reframes privilege: when an admin’s token is stolen at 2 p.m. (Lesson 1.1’s test), what does the attacker hold? In the standing-privilege world: Global Admin, forever, everywhere. In the world this lesson builds: whatever was activated right then — usually nothing.
The mechanism is just-in-time elevation: privileged roles are held as eligible, not active. An eligible admin is a normal user until they activate — a gated ceremony (phishing-resistant MFA, business justification, optionally approval for the crown-jewel roles) that grants the role for a bounded window, hours not months, logged end to end.
Standing world: identity IS the privilege - steal it, own everythingJIT world: identity MAY REQUEST privilege - steal it, and the attacker must perform the activation ceremony: strong-auth gate, justification on the record, alert fired, clock tickingPrompt first
Switch your workspace agent into plan mode on your role-assignment export and start with (if you do not have live exports or files handy, ask it to generate a small Meridian practice sample from the lesson’s scenario first):
Plan our conversion from standing privileged roles to just-in-time eligibleassignments across directory, cloud, and SaaS consoles, including activationgates, quarterly reviews, and workstation tiers. Ask clarifying questionsabout our tier-0 boundaries first.Check that the plan or draft covers:
- Complete role inventory across directory, cloud, and SaaS admin consoles showing current assignment counts and standing-to-eligible conversion targets
- Per-tier activation gates (phishing-resistant MFA always; approval plus max 4-hour window for tier-0; 8-hour window for tier-1), leaving only the two break-glass accounts standing and alarmed
- Quarterly access reviews owned by the role owner (not the admin’s manager), with an automatic removal flag for any eligibility unused across two consecutive quarters
- Tiered workstation rules requiring a hardened privileged access workstation or cloud PC for tier-0 roles so credentials never touch a daily-driver browser profile
If the draft assigns reviews to line managers or leaves long activation windows, follow up with:
Route quarterly reviews to each role's service owner, enforce the two-quarterinactivity removal rule, and cap tier-0 activation at four hours.Three design points that make the model hold:
The activation gate is a PDP moment. Activation is itself an access decision — evaluated with the strongest signals: phishing-resistant strength (Lesson 2.2), compliant device (Module 3), and for tier-0, a second human. This is where the pillars you’re building compound: the gate is only as strong as the weakest signal it accepts.
Reviews prune by evidence, not memory. Quarterly reviews fail when they’re a rubber-stamp email to managers. The version that works reviews activation logs: a role never activated in six months is a role that person doesn’t need — removal is the default and retention needs the justification. (The same inactivity logic that Lesson 6.1 applies to groups and projects.)
The tier boundary is physical. Tier-0 (identity infrastructure itself) is administered only from hardened, separately-managed workstations — because a domain admin activating from the same browser profile that opens lunch-menu links makes the JIT ceremony theater. The boundary between “daily driver” and “admin surface” is the one air gap left that’s worth its cost.
Admins experience JIT as friction — “I activate five times a week!” Two honest answers. First: five activations a week means five moments an attacker must also survive, each one alarmed; that IS the control. Second: streamline what’s streamlinable — activation approval only on tier-0, sensible window lengths, no re-auth mid-window. The goal is standing privilege at zero, not activation ceremonies at maximum. If someone activates daily for a routine task, that task probably deserves a scoped automation identity (next lesson) instead of a human admin role.
A quarterly review shows a database engineer has held eligible 'Subscription Owner' for a year and activated it exactly twice — both during a migration that ended in March. The engineer objects to removal: 'I might need it, and re-granting takes a day.' What does the model say?
Key takeaway
Privilege that stands is privilege pre-stolen. Make every role eligible, gate activation with your strongest signals, bound it in hours, review by activation evidence with removal as the default, and put tier-0 behind hardware you don’t browse on. The 2 p.m. token thief now holds a normal user and an alarm bell. One identity class left — the one that outnumbers everyone: the machines.