Skip to content
Tech Interview Prep
Technical interview guide

Privileged Access Management

Extra controls around high-risk, elevated-privilege accounts: the accounts that matter most if compromised.

Read
45 min
Practice MCQs
25
Interview Q&A
25
Edition
v5
Editorial status
Reviewed

Scope: NIST SP 800-53 Rev. 5 and SP 800-207, NIST SP 800-63B-4, current CISA, AWS, Microsoft Entra PIM, Google Cloud PAM, Kubernetes, IETF, OWASP, and sudo guidance reviewed 2026-09-04.

Curated: · Written: · Reviewed:

25 questions

MCQ-3An engineer's laptop is compromised by a malicious browser extension that reads cookies from the browser profile it is installed in. The engineer's daily identity (email, Slack, code review) is signed in in the default profile. The privileged identity (production console, secrets manager) is signed in only in a separate browser profile that is opened for admin work and closed when finished; its session cookie is not written to disk. The extension is installed in the default profile. What does the attacker obtain?
MCQ-15Acme's payments team has a 4-hour maintenance window. Two engineers from support vendor Vantage must run `ALTER TABLE` on the `payments` database. Vantage's IdP supports SAML but not SCIM. The current process gives all Vantage staff one shared VPN credential and one permanent `prod-admin` role. After the window, the audit log shows the `ALTER TABLE` succeeded but attributes it to `prod-admin`, not to either engineer. Which control would have made the action attributable to the individual engineer without blocking the window?
MCQ-1A 30-person production admin team moves to just-in-time access: each admin requests a 2-hour grant about five times a week, every request in the log carries a named approver, and the grant volume matches that assumption. Six months later the audit measures 5,040 admin-hours per week of continuous elevated access across those 30 human identities: the audit counts only those identities, not service accounts. Which finding explains the measurement?
MCQ-22A bank's identity team replaced permanent admin role assignments with a just-in-time approval flow. Two quarters in, the CISO wants one metric that shows standing privilege actually shrank. The team can query the IAM system for every principal's effective permissions at any moment. Which metric should they report?
MCQ-2A team already grants time-boxed elevation: an engineer requests Domain Admin for a two-hour window and the role is revoked automatically at expiry. A review finds that during those windows the engineer can still run any command against any object in the domain. Which change adds just-enough administration to the existing process?
MCQ-16A quarterly privileged access review exports the PIM eligible-assignments report for a production subscription: 412 rows across 96 accounts. The team revokes the 180 rows whose accounts show no sign-in in 90 days, then closes the review. The next audit finds 37 accounts with Owner on the subscription that the review never examined. Twelve hold Owner as a direct assignment made outside PIM; the other 25 are members of a group that holds Owner directly, so the export lists the group, not the people. Which change to the review would have surfaced those 37 accounts?
MCQ-25An organization must maintain emergency break-glass root access to its cloud infrastructure in case enterprise single sign-on (SSO) and federation fail. Which architectural control pattern best secures the break-glass credentials while ensuring strict auditability and non-repudiation during an incident?
MCQ-9A payments team's break-glass credential for the orders DB is a sealed envelope in a data-center safe, scoped to read-only on orders DB. Opening the safe pages the security team. A rotation job on a bastion runs 5 minutes after each retrieval: it logs in to the DB with a service account whose authentication goes through the corporate IdP, then changes the credential's password. Okta goes down at 02:00. At 02:05 an SRE opens the safe, retrieves the credential, and runs the read-only query; it succeeds. The SRE stays on the incident and does not return the credential to the safe, expecting the rotation job to invalidate it at 02:10. Okta recovers at 02:40. What is the state of the break-glass credential at 03:00?
MCQ-4During a Sev-1 on the payments database, on-call DBA A requests a 2-hour break-glass admin role to run the DDL that finishes a half-applied migration; read-only access cannot complete it. The approval policy requires an approver who is not the requester, not on the requester's on-call rotation, and not the incident commander. Three people hold approval rights: B (A's teammate, same rotation, already paged into the incident), C (the incident commander, who opened the bridge and requested the elevation on A's behalf), and D (a staff engineer in another org, not on call, who has never worked on payments). The break-glass approval SLA is 15 minutes. What should the approval workflow do?
MCQ-23A 300-account AWS Organization runs a quarterly access review. CloudTrail for the management account shows 9 root-user events in the last 90 days: 6 are `CreateServiceLinkedRole` calls at 02:00 UTC from a CI pipeline that reads root credentials out of a secrets manager, 2 are console sign-ins from one engineer's home IP, and 1 is a `PutAccountSetting` that changed the account contact email. Root MFA is enabled and the root user has no access keys. Which single change removes the largest number of those root events?
MCQ-6An SRE team moves the AWS access key for its deployment IAM user into a vault. The key is rotated weekly, and only members of the `deploy` group can read it. Over one quarter, CloudTrail records 2,140 API calls made with that key, every one attributed to the single IAM user `deploy-bot`, and the IAM policy attached to that user is `AdministratorAccess`. Which conclusion does this evidence support?
MCQ-24An administrator's only route into AWS is Okta SSO: they sign in to Okta, open the AWS tile, and assume the `AdminOps` role via SAML with a 3600-second session. At 09:40 on their last day they sign in and export the resulting credentials into a shell. At 10:00 offboarding disables their Okta account and removes the AWS app assignment. At 10:05, which statement about those exported credentials is correct?
MCQ-19CloudTrail shows the role `deploy-prod` was assumed via `sts:AssumeRole` at 02:14 UTC from 203.0.113.47, an address outside your CI runner's range; the call was signed with the CI runner's long-term access key. The session token issued has a 12-hour lifetime, and it is now 02:40 UTC. The role reads `prod/db/*` from Secrets Manager and pushes images to ECR. Which single action stops the attacker's current session from making further API calls?
MCQ-11A PAM broker grants a 15-minute elevation to a production database. At approval, 09:00, it mints a session token with a 12-hour TTL and hands it to the client. The database validates the token's signature and expiry at connection setup and has no network path back to the broker. The broker's revocation loop runs every 5 minutes and deletes the grant record once the elevation window closes. An engineer approved at 09:00 keeps the token. Until when can they open a new database connection with it?
MCQ-5A JIT elevation service issues one credential per engineer per day: a 24-hour TTL, accepted by every service in the cluster. An engineer activates the role at 02:00 to read one row from the production database, finishes at 02:15, and the credential is never revoked. At 06:00 the same credential is replayed from a compromised CI runner against the secrets manager and succeeds. Which single change to the issuance policy blocks the 06:00 replay without affecting the 02:00 task?
MCQ-12A payments platform runs 60 services on Kubernetes 1.29. Each pod mounts a projected service-account token (expirationSeconds 3600, audience set to the cloud STS endpoint) and assumes an IAM role through OIDC federation. Fifty-nine services use the cloud SDK's default credential chain and run for weeks without auth errors. The sixtieth reads the token file once during startup, stores the string in a struct field, and reuses it for every call; its pods begin failing with an STS `InvalidIdentityToken` error about 60 minutes after each start, and restarting a pod clears the error for another hour. Which change removes the recurring failure?
MCQ-17An admin API requires acr = urn:mfa and auth_time within 300 seconds for POST /v1/clusters/{id}/rotate-keys. Alice signed into the console at 09:00 with password + TOTP. At 13:00, from that same console session, she calls rotate-keys. At 12:55 she had completed a TOTP challenge in a separate session on her phone. What should the gateway do?
MCQ-20A support console lets agents act on customer accounts through OAuth 2.0 token exchange (RFC 8693). The exchanged token carries `sub` = customer C, `act` = {sub: agent A}, and `aud` = "billing-api". billing-api authorizes refunds by looking up the token's subject in the customer table and applying that customer's tier limit; C's tier limit is $500. The IAM system separately caps any single refund made by an agent role at $50. Agent A submits a $400 refund for C, and billing-api approves it. The request log shows the token arrived with both `sub` and `act` intact. Which statement explains the approval?
MCQ-13A secrets broker runs as a service account holding `db_admin` on every shard. It verifies each caller's OIDC token and that the caller holds at least one active grant, then opens a session to the target named in the request body. An audit finds a developer whose only grant covers shard A opened a session to shard C, and the broker logged the session as authorized. Which flaw produced this?
MCQ-14A jump broker enforces a production command allowlist. It checks that the client's command line starts with an allowed binary, then forwards the line and the client's environment to the target. Three engineers reached an interactive shell on production last month; in each case the broker's log shows the command line was exactly `kubectl get pods` and passed the allowlist, and the target's process tree shows the shell was a child of `kubectl`, not a second command. Which change closes this?
MCQ-7Twelve Solaris 10 hosts run a legacy inventory job that accepts only the shared root account; 40 operators reach them through a privileged access management broker that authenticates each operator with SSO. The target audit logs record the shared account, the command, and the timestamp, but the legacy audit configuration cannot be changed to add fields, and an operator may have cleared the shell history. An auditor must map a specific rm -rf /data at 02:14 to the individual operator. Which change makes that mapping possible?
MCQ-18A PAM connector rotates the local administrator password on 2,400 Windows servers. It runs 8 workers in parallel, and each target takes 4 seconds to rotate. The vault issues the connector a single credential with a 15-minute TTL, and the connector cannot renew it. What happens on a full run?
MCQ-8A bastion host records every privileged SSH session to WORM object storage with 90-day retention. During one session an engineer pastes a production database password into a psql prompt, then reads a customer's account record. Audit needs the recording to show who ran which command against which host. What should the recording pipeline do?
MCQ-21A bastion host records every privileged SSH session to a local spool and uploads it to object storage asynchronously. Policy requires that every privileged session in the PCI production fleet be recorded with no unrecorded window, and that any recorder failure be visible to the on-call security engineer within one minute. The corporate fleet that shares the bastion is out of scope for the fail-closed requirement, and its sessions must not be terminated by a recorder failure. At 02:14 the uploader's object-storage credential expires and uploads start failing; the 8 GB spool was 6 GB full at 02:14 and the sessions produce 2 MB/s in aggregate, so at 02:31 the recorder's next write returns ENOSPC. What should the recorder do?
MCQ-10A PAM vault writes every audit event to a local append-only segment file and forwards events to a SIEM. The forwarder runs once every 60 seconds and sends everything generated since its last successful run. While it is disabled it buffers nothing: events generated during the outage are dropped and never sent. The SIEM is append-only, retains 400 days, and no vault role can write to it. mallory holds two roles: PolicyAdmin, which can edit safe policies and the SIEM forwarder configuration, and LogAdmin, which can delete local segment files. On one day she disables the forwarder at 13:20. The vault records the change and the forwarder sends it in its final run before stopping. Checks out the domain-admin credential from safe PROD-DA at 13:58, deletes the local segments covering 13:00–14:00 at 14:02, and re-enables the forwarder at 14:05. Which of these four actions appear in the SIEM's copy of the audit log?