The interview question that trips up 70% of candidates
Final-round interview: "In CyberArk Privilege Cloud, what does CyberArk run and what do you run?"
Wrong answers: "CyberArk runs everything", "You install the Vault on AWS yourself". Right answer: CyberArk hosts the Digital Vault, the PVWA web portal, the backend, the encryption keys, the upgrades and the HA/DR on AWS across three availability zones. You run only one thing on-premises: the Connector — a server bundle holding CPM and PSM that reaches your target systems and talks outbound to CyberArk over TCP 443. Knowing that exact line lets you scope a migration, design firewall rules, and explain a 99.95% SLA to a nervous CISO who thinks SaaS means "we lost control".
💡 The piped city gas analogy
Self-hosted PAM is the LPG cylinder in your kitchen. You own it, you check the pressure, you call the dealer when it runs out, and if the regulator leaks at 2am, that is your problem. Privilege Cloud is piped city gas. The supply company (CyberArk) owns the plant, the pipeline and the pressure. You manage only the burner and the pipe inside your home — the Connector running CPM and PSM. No more midnight Vault patching. But you also cannot install a custom valve the company has not certified, and if the pipeline goes down, you depend on them to bring it back. That trade — less daily hassle for less raw control — is the whole story of this blog.
4 things you'll be tested on before we begin
The only on-prem box in Privilege Cloud. Holds CPM (rotation) + PSM (session proxy). Talks outbound to CyberArk over TCP 443 via the Secure Tunnel. So what: firewall rules and uptime here are entirely yours.
Legacy CPM is active-passive, XML-configured, single-server. SRS (Secrets Rotation Service) is active-active, UI-driven, SaaS-hosted, auto-scaled. So what: SRS removes the single-CPM split-brain risk and centralizes logging.
Emergency admin account on a "No Change" platform, with an offline sealed copy under dual custody. Used only when the Vault is unreachable. So what: a rotated break-glass account is the #1 cause of being locked out during an outage.
Defender (PAM-DEF) = daily ops. Sentry (PAM-SEN) = install + advanced config. Guardian = strategic architecture. CDE = partner-only delivery track. So what: each maps to an interview round — screening, technical, architecture.
① Privilege Cloud vs Self-Hosted — the responsibility split
In self-hosted PAM you own everything: the Vault servers, the SAN, the cluster, the patching, the DR replication. In Privilege Cloud, CyberArk hosts the Vault, the PVWA portal (at <org>.privilegecloud.cyberark.cloud), the backend on AWS across 3 AZs, the per-customer encryption keys, the upgrades and the HA/DR. You keep only the Connector — and your target-side firewall rules, your network, and your break-glass offline copies.
Priya Sharma is CISO at NeoFinance Bank. They run CyberArk PAM Self-Hosted v12.1 on ageing DC1 hardware. Two engineers babysit a Vault cluster for 1,200 accounts — and burn 40% of their week on patching, Windows updates and HA maintenance. Twelve months brought three Vault incidents: a split-brain cluster event, a failed upgrade rollback, and a CPM sync break after a Windows patch.
The fix Priya signs off: migrate to Privilege Cloud, hand the Vault-ops burden to CyberArk, and keep her two engineers on onboarding and policy — not on SAN babysitting. RBI's 5-year audit-retention mandate is met by Privilege Cloud's built-in syslog-to-SIEM integration.
Gain: no Vault hardware, no SAN, no cluster nights, a CyberArk-managed 99.95% SLA, auto-upgrades, and SRS active-active rotation. Give up: the ability to air-gap the Vault (the Connector needs outbound internet), some custom-platform behaviours that differ on SaaS, and a hard dependency on CyberArk cloud uptime for any credential retrieval. For a two-engineer bank team, the trade is almost always worth it.
Pause & Predict: NeoFinance's security architect insists "we must be able to retrieve a credential even if our internet is down". Does Privilege Cloud satisfy that?
Pause & Predict In Privilege Cloud, a CISO asks you to draw a single box around "everything CyberArk runs" and another box around "everything we run". What goes in each box?
② The Identity Security Platform — one admin plane
Privilege Cloud does not live alone. It sits inside the Identity Security Platform Shared Services (ISPSS) — a single SaaS layer that gives you unified Identity Administration, SSO/MFA via CyberArk Identity, centralized Audit, Identity Security Intelligence (UBA feed to your SIEM), and Secure Cloud Access. One admin console fronts Privilege Cloud, EPM, Conjur/Secrets Manager and CyberArk Identity together — so you stop logging into five disconnected portals.
Arjun Mehta is a senior PAM engineer deploying CyberArk for PharmaCorp (FDA 21 CFR Part 11 + CDSCO regulated). Before ISPSS, his admins juggled separate consoles for vaulting, endpoint privilege and secrets. Now one Shared Services tenant gives a single sign-on, one MFA policy and one audit feed across all of it.
The lesson he teaches juniors: the Platform is not a marketing wrapper. The shared identity directory and unified audit are what let a regulator trace one user across Vault retrieval, endpoint elevation and secret access in a single timeline. That single timeline is the compliance win.
SRS — the rotation engine that replaces CPM
In self-hosted PAM, the CPM rotates passwords. It is active-passive and you cannot load-balance it — two CPMs hitting the same account rotate to two different values and the account becomes inaccessible. In Privilege Cloud the default engine is SRS: accounts bind to an SRS connector pool (not a single server), giving active-active HA for rotation with distributed locking, and centralized logging in the CyberArk dashboard instead of scattered per-server log files. Both can coexist during migration.
$pvwa = "https://neofinancebank.privilegecloud.cyberark.cloud/PasswordVault"
$body = @{ username = "svc_api_admin"
password = (Read-Host -AsSecureString "PAM API pwd" |
ConvertFrom-SecureString -AsPlainText) } | ConvertTo-Json
$token = Invoke-RestMethod -Uri "$pvwa/API/auth/Cyberark/Logon" `
-Method POST -Body $body -ContentType "application/json"
$hdr = @{ Authorization = $token }
Write-Output "Authenticated. Token: $($token.Substring(0,20))..."
Authenticated. Token: eyJhbGciOiJSUzI1NiIs... PS C:\> $hdr.Authorization.Length 612
Arjun's team wants HA for credential rotation in Privilege Cloud. A junior suggests "just run two CPM servers behind a load balancer". Why is this wrong, and what is the right answer?
Recreated for clarity☁️ The exact screen you'll open every morning — ISP Portal → Dashboard. Your tenant matches this layout.
③ The go-live playbook — discover, onboard, isolate, detect, optimize
A CyberArk go-live is not "install and switch on". The proven sequence is discover → onboard → isolate → detect → optimize, mapped to CyberArk's own Rapid Risk Reduction order: secure IT-admin credentials first, then stop credential theft (EPM), then protect DevOps secrets (Conjur), then detect threats (PTA). Discovery uses the accounts feed to populate the Pending Accounts list; onboarding rules auto-vault what matches and leave the rest for manual review.
PharmaCorp onboarded 8,000 accounts over a single weekend because the CISO wanted "numbers for the board". Sixteen Oracle DB and 23 Windows service accounts went in with no dependency discovery. CPM rotated them immediately (default ChangeNotificationOnlyOnExpiry=No). By Day 3, seven overnight ETL jobs feeding the manufacturing execution system failed with Event ID 4625 logon failures — scheduled tasks and ODBC DSNs were still using the pre-rotation passwords. Production halted for 6 hours.
Fix: always run Dependency Discovery (PVWA → Account Details → Dependencies → Discover) before onboarding service accounts, set ChangeNotificationOnlyOnExpiry=Yes for a 30-day stabilization window, and cap go-live velocity at 200 accounts/week with an app-owner sign-off gate.
$pvwa = "https://neofinancebank.privilegecloud.cyberark.cloud/PasswordVault"
$acct = @{
name = "prod-db01-sa_svc_oracle"
address = "10.10.5.25"
userName = "sa_svc_oracle"
platformId = "OracleDatabase"
safeName = "PROD-DB-ServiceAccounts"
secret = (Read-Host -AsSecureString "Current pwd" |
ConvertFrom-SecureString -AsPlainText)
platformAccountProperties = @{ Port = "1521"; Database = "PRODDB" }
} | ConvertTo-Json
$r = Invoke-RestMethod -Uri "$pvwa/API/Accounts" -Method POST `
-Headers $hdr -Body $acct -ContentType "application/json"
Write-Output "Account onboarded. ID: $($r.id)"
Account onboarded. ID: 453_6 Dependencies found: 3 Scheduled Tasks, 2 IIS App Pools Rotation held: ChangeNotificationOnlyOnExpiry=Yes (30-day window)
Pause & Predict: NeoFinance onboards their break-glass firewall-admin account with the standard Windows platform to "keep it consistent". Two weeks later their tenant is unreachable in an ISP outage. What happens when they open the sealed envelope?
HA, DR and the SLA you can promise
Privilege Cloud carries a 99.95% availability SLA. Within a region, HA spans 3 AWS availability zones, so most failovers are near-zero to ~20 minutes RTO. Cross-Region DR (CRDR) is an add-on license; without it, a full DR backup-restore has a 4–24 hour RTO. In self-hosted, you carry all of this yourself: a two-node Vault cluster on shared SAN (SCSI-3 Persistent Reservations), a Cluster Vault Manager watching health, a virtual IP for failover, and a quorum disk to stop split-brain.
An enterprise ran a two-node HA Vault with the quorum disk "temporarily" removed during a storage migration three months earlier — and never reinstalled. A 90-second heartbeat brownout at peak hours let both nodes declare themselves active, seize the virtual IP and write to shared SAN concurrently. Result: partial Safe corruption and a 6-hour DR restore. Lesson: never run a two-node cluster without a quorum disk, and drill failover quarterly. This is precisely the class of pain Privilege Cloud removes by hosting the Vault for you.
PharmaCorp's PSM sessions establish from the portal but drop after exactly 30 seconds. The Connector sits behind a next-gen firewall with TLS inspection enabled, and firewall logs show "allow" on port 443. What is the cause and fix?
Recreated for clarity🔌 The exact screen you'll check when "PSM won't launch" — Administration → Connector Management. Your tenant matches this layout.
| Connector | Components | Version | Status |
|---|---|---|---|
| connector-mum-01 | PSM, CPM, Secure Tunnel | 14.2 | Connected |
| connector-mum-02 | PSM, CPM | 14.2 | Connected |
| connector-blr-01 | CPM | 14.0 | Connected |
| connector-blr-02 | PSM | 14.2 | Disconnected |
Pause & Predict NeoFinance runs the proven go-live sequence. A junior wants to onboard the break-glass firewall-admin account on day one "to get it out of the way". In the discover → onboard → isolate → detect → optimize order, when should break-glass actually go in, and why?
During NeoFinance's go-live you onboard a Windows service account that drives 3 scheduled tasks and 2 IIS app pools. What is the correct sequence to avoid a production outage when CPM first rotates it?
Event ID 4625; (c) breaks your rotation policy; (d) is the onboard-too-fast pitfall.The certification ladder — and the interview round each one is
CyberArk's ladder is three public rungs plus a partner track. Defender (PAM-DEF) tests daily operations and Vault admin. Sentry (PAM-SEN) tests deployment, install, advanced config and troubleshooting. Guardian tests strategic, multi-solution architecture. CDE (Certified Delivery Engineer) is partner-only for implementation consultants. All public exams run via Pearson VUE, roughly $200, 90 minutes, multiple-choice plus scenarios.
Karthik is interviewing for a CyberArk role. Round 1 (screening) is Defender-level: explain the Vault, CPM and PSM, name the Vault protocol port, describe what happens when a session launches. Round 2 (technical) is Sentry-level: troubleshoot a PSMSC036E AppLocker block, design an HA cluster, explain CPM split-brain. Round 3 (architecture) is Guardian-level: design a Privilege Cloud migration for a 5,000-account bank and map the Blueprint to a kill chain.
The takeaway he gives juniors: study to the rung you are interviewing for. Reading Guardian architecture before you can answer "what port does the Vault use?" wastes everyone's time.
In a Sentry-level (PAM-SEN) interview at a Mumbai SI, you are asked: after a PSM v14 upgrade, sessions to a legacy SCADA HMI fail with error PSMSC036E. What is the diagnosis and fix?
PSMSC036E is the AppLocker-block signature. PSM hardening whitelists known binaries; a custom connection component's executables get blocked after an upgrade reapplies policy. The fix is to read the AppLocker audit log and add the component binaries to the whitelist (hash rules for unsigned ones). (a/c/d) target unrelated subsystems.④ The capstone — replay a breach, break each stage with CyberArk
Here is the whole series tied into one scenario. This kill chain is the canonical 2022 privileged-credential breach pattern: an attacker phishes an IT admin, drops on the endpoint as local admin, dumps cached credentials, finds hardcoded PAM admin creds in a script, moves laterally, forges a golden ticket, and reaches domain dominance. For each stage, one specific CyberArk control breaks the chain.
Priya's board asks the only question that matters: "If we are breached like the headlines, where does CyberArk actually stop it?" She walks them stage by stage — phish, local admin, credential dump, hardcoded-secret discovery, lateral move, golden ticket — and names the control that breaks each one: Identity MFA, EPM, vaulting + rotation, Conjur, JIT/Zero Standing Privilege, and PTA.
Defense-in-depth means the attacker does not need to beat one control — they need to beat all six in sequence, which is why a layered CyberArk deployment turns a domain-dominance breach into a contained alert.
▶ Watch the kill chain meet each control
Replay the breach against a fully-deployed CyberArk fleet. Each stage hits a wall. Press Play.
The 2022 Uber incident ran almost this exact chain: vishing → MFA fatigue/reset → a PowerShell script with hardcoded PAM-admin credentials on an internal share → full PAM access → every organizational secret. The five CyberArk controls that would have broken it: Conjur (no hardcoded creds), PSM session isolation (admins never see their own password), Identity MFA (blocks the vishing reset), JIT (time-limited admin), and PTA (mass-credential-access alert fires immediately).
Pause & Predict: a junior argues "if we just deploy EPM to kill local admin, we don't need vaulting or PTA". Where does that argument break?
Pause & Predict In the capstone kill chain the attacker reaches the final stage — a forged golden ticket for domain dominance. Every earlier control (Identity MFA, EPM, Conjur, JIT) has already been beaten. Which single CyberArk control is purpose-built to catch this last stage, and what does it do?
🤖 Ask the AI Tutor
Tap any question — instant context-aware answer.
Deeper questions → chat.techclick.in.
The classic go-live pitfalls — and the one-line fix for each
8,000 accounts in a weekend, no dependency map, CPM rotates immediately, production scheduled jobs die. Cap velocity at 200/week, run Dependency Discovery first, hold rotation with ChangeNotificationOnlyOnExpiry=Yes.
Tenant unreachable in an outage and the only admin credential is in the Vault — or the sealed copy is stale. Use a "No Change" platform + refreshed offline copy + dual custody.
One service account runs three scheduled tasks and two IIS app pools; rotation breaks all five. Always map dependencies before enabling management so CPM updates them on rotation.
Dozens of custom connection components, untested on SaaS PSM, break after an upgrade with PSMSC036E (AppLocker) or PSMSR037E (Vault comms). Re-validate every custom component in the SaaS tenant before cutover.
A TLS-inspection firewall re-signs the Secure Tunnel cert and silently drops the connection. Add the CyberArk cloud endpoints to the DPI bypass list.
One last security note — patch the CVEs
Even SaaS-adjacent components need patch hygiene. CVE-2024-54840 (CVSS 4.2) is a PVWA Host-header injection enabling open-redirect/phishing in PAM Self-Hosted before 14.4 — upgrade to 14.4+. The July 2025 Conjur/Secrets Manager set is more serious: CVE-2025-49827 and CVE-2025-49831 (both CVSS 9.1) are IAM authenticator bypasses, CVE-2025-49828 (CVSS 8.6) is an RCE, and CVE-2025-49830 (CVSS 7.1) is a path traversal. Patches are on the CyberArk Marketplace — apply within your KEV-aligned window.
📝 Check your understanding — 10 questions, 70% to pass
Q1–Q3 above already count. Below are Q4 to Q10.
In CyberArk PAM Self-Hosted, which dedicated TCP port do all internal components (PVWA, CPM, PSM, PTA, PACLI) use to talk to the Digital Vault?
You are onboarding a Windows service account at NeoFinance that runs 3 scheduled tasks and 2 IIS app pools. What is the correct order to avoid a production outage?
NeoFinance's Privilege Cloud tenant is unreachable in an ISP outage and the SOC needs a critical firewall-admin account. Judging the four options below, which one is the correct, audit-defensible break-glass procedure for a well-designed deployment?
A self-hosted HA Vault cluster's active node loses its PrivateArk Server service. Which condition would PREVENT automatic failover to the passive node?
In the capstone kill chain, the attacker finds a PowerShell script on a network share that previously held hardcoded PAM-admin credentials. Which CyberArk control specifically breaks this stage?
In the capstone, the attacker forges a golden ticket to reach domain dominance. Which control detects and contains this final stage, and how?
Priya's board asks whether the 5,000-endpoint migration from self-hosted to Privilege Cloud is the right call. What is the most complete senior-engineer evaluation?
"Privilege Cloud is PAM as piped gas: CyberArk runs the Vault, you run just the Connector — and a real breach is broken stage by stage by MFA, EPM, vaulting, Conjur, JIT and PTA, not by any single magic control."
Before you leave: in your own words, explain to an imaginary junior why a break-glass account must be on a "No Change" platform, and which two CyberArk controls break the endpoint stages of the kill chain. If you can say both without scrolling up, the series has landed. Want a spaced reminder? The end-of-lesson tracker will resurface these in 7 days when you revisit.
Glossary — the terms that came up
- Privilege Cloud
- SaaS-delivered CyberArk PAM. CyberArk hosts the Vault, PVWA and backend on AWS; you run only the on-prem Connector. Tenant URL:
<org>.privilegecloud.cyberark.cloud. - Connector
- The single on-prem box in Privilege Cloud, bundling CPM (rotation) and PSM (session proxy). Talks outbound to CyberArk over TCP 443 through the Secure Tunnel.
- SRS (Secrets Rotation Service)
- Cloud-native, active-active rotation engine replacing legacy CPM. UI-configured, auto-scaled, centrally logged; accounts bind to an SRS connector pool.
- ISPSS / Identity Security Platform
- The unified SaaS layer — Identity Admin, SSO/MFA, Audit, Security Intelligence, Secure Cloud Access — that sits above Privilege Cloud, EPM, Conjur and SCA.
- Break-glass account
- Emergency admin account on a "No Change" platform with an offline sealed copy under dual custody, used only when normal PAM access is unavailable.
- PTA (Privileged Threat Analytics)
- Behavioral + DPI engine that detects golden tickets, Pass-the-Hash and anomalous Vault access in real time, assigning risk scores and enabling auto-suspension.
- EPM (Endpoint Privilege Manager)
- Removes local admin rights from endpoints and blocks credential theft from OS stores and browsers; supports JIT elevation.
- Quorum disk
- The tiebreaker voter in a self-hosted HA Vault cluster that prevents split-brain when the heartbeat network partitions.
🎓 You've completed the CyberArk series — now prove it
Ten blogs done: from the Vault and CPM/PSM internals to Privilege Cloud, the Identity Security Platform, go-live, and the kill-chain capstone. The last step is the practice set — full Defender/Sentry-tier MCQs under exam conditions, scored.
Sources cited inline
- CyberArk Docs — Privilege Cloud Connector system requirements
- CyberArk Docs — Secrets Rotation Service (SRS) architecture
- CyberArk Community — Self-Hosted to Privilege Cloud migration checklist
- CyberArk Product Security — July 2025 Conjur/Secrets Manager CVEs
- GitHub Advisory — CVE-2024-54840 (PVWA Host header injection)
- Pearson VUE — CyberArk certification program (Defender/Sentry/Guardian)
- CyberArk — EPM credential-theft blocking