Skip to content
Tech Interview Prep
Technical interview guide

Zero Trust Architecture

Verifying every request explicitly instead of trusting anything by network location.

Read
29 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: NIST SP 800-207; NIST SP 800-207A; NIST SP 1800-35 final (June 2025); CISA Zero Trust Maturity Model 2.0.

Curated: · Written: · Reviewed:

25 questions

MCQ-4A zero-trust gateway receives a request from a contractor's laptop to reach the finance database. The policy engine evaluates the request and returns 'allow', specifying that the session must use a short-lived credential. Which action does the policy administrator take next?
MCQ-5Two services guard the internal HR API. One evaluates device posture and user attributes and returns allow or deny with a 15-minute session limit. The other sits inline on the request path. Which responsibility belongs to the inline service?
MCQ-2A zero trust policy engine evaluates each request at session start and re-evaluates every 5 minutes. At 09:00 it grants a user a session to a sensitive repository: identity verified with a FIDO2 key, device managed and compliant, connection over the corporate network. At 09:35 the EDR agent reports the device has loaded a kernel module on the blocklist; the user is 60% through downloading a 4 GB archive. What should the engine do at the 09:35 re-evaluation?
MCQ-6A company's authorization service grants admin scope to any request carrying the `network=internal` tag that its VPN concentrator adds. An attacker stole a regular employee's VPN session token, connected through the VPN, and received admin scope. Which change fixes the flaw that let this happen?
MCQ-19A partner's IdP issues OIDC access tokens for two OAuth clients: `partner-intranet` (the partner's intranet portal) and `partner-api` (the partner's service accounts). This IdP sets `aud` to the requesting client's ID, not to the resource server. Your gateway validates the signature against the IdP's JWKS, checks `iss == "https://idp.partner.example"`, and authorizes on `scope` containing `payments:write`. A pen test shows a partner contractor with only intranet access can call `POST /v1/payments` on your API. The contractor's token is `{iss: "https://idp.partner.example", aud: "partner-intranet", scope: "payments:write", sub: "u-contractor"}`. The partner's three service accounts (`svc-payments`, `svc-batch`, `svc-recon`) call the same endpoint with tokens like `{iss: "https://idp.partner.example", aud: "partner-api", scope: "payments:write", sub: "svc-payments"}` and must keep working. Which single gateway change closes the hole?
MCQ-1Northwind's internal admin console is reachable at 10.4.2.11 by anyone who completes the corporate VPN login. The security team puts an identity-aware proxy in front of it at admin.northwind.internal; the proxy checks the caller's SSO group and device posture on every request and forwards approved traffic to 10.4.2.11. An audit finds that a contractor with a valid VPN session but no admin group reached the console by connecting directly to 10.4.2.11. Which change closes that gap?
MCQ-8A mesh issues every workload a SPIFFE SVID with a 1-hour TTL, and each sidecar renews its workload's SVID in place. A nightly batch Job runs for 6 hours with the sidecar disabled and loads its SVID once from a file at startup. Its calls to the ledger API succeed from 02:00 to 03:00, then every call fails the mTLS handshake for the rest of the run. Always-on services calling the same ledger API are unaffected, and no CA, trust-bundle, or authorization-policy change was made that night. What is the cause?
MCQ-20A contractor signs in from a personal laptop that has no EDR agent and cannot attest disk encryption. The user's identity is verified with MFA, and the request is for the internal payroll admin console. The zero-trust policy engine evaluates device posture and resource sensitivity. What should the engine do?
MCQ-9A zero-trust gateway checks device posture at 09:00 (disk encryption on, EDR agent reporting, OS patch level current) and the user signs in through SSO with MFA at the same time. At 09:40 that session calls GET /finance/payroll/export.csv. The gateway's baseline policy lets compliant devices reach internal apps; the finance app's own policy requires an MFA within the last 15 minutes for any export action. What does the gateway do with the request?
MCQ-10A payments service on an EC2 instance reads objects under one S3 prefix and writes to one DynamoDB table. It authenticates with an IAM user's long-lived access key stored in a config file that four teams can read. An audit flags the key as standing privilege. Which change most reduces the standing privilege the key grants?
MCQ-15A SaaS billing platform runs 30 services behind an identity-aware gateway that validates each JWT's signature, expiry and `tid` (tenant) claim, and a mesh that enforces mTLS between services. A pen test authenticated as a user in tenant 4471 and fetched `GET /invoices/88213`, which belongs to tenant 9902; the invoice service returned HTTP 200 with the full record. The handler runs `SELECT * FROM invoices WHERE invoice_id = $1`. Which change closes the hole?
MCQ-14`ledger` runs in a mesh that enforces STRICT mTLS on every inbound connection, and every request arrives with a verified SPIFFE ID. Over six minutes, a compromised `frontend` pod (identity `spiffe://cluster.local/ns/shop/sa/frontend`, certificate valid and unexpired) sent 47 `POST /admin/rotate-key` calls to `ledger`; all 47 returned 200 and the signing key rotated each time. In the same window `frontend` made 1,204 `GET /v1/read` calls that must keep working. Which change would have blocked the 47 admin calls and left the 1,204 reads working?
MCQ-7A platform team already runs mTLS and a per-request authorization policy on every service call. A post-incident review found that a compromised web pod could open TCP connections to 380 other workloads and reached the cardholder database through one of them. The team is adding microsegmentation. Which statement should the design doc make about what the new layer contributes?
MCQ-25An internal admin console sits behind a zero trust gateway. Every request must carry an access token from the identity provider; the gateway asks the policy engine for an allow/deny decision and caches that decision for 60 seconds, keyed by user and resource. The gateway does not subscribe to revocation events: only the policy engine does. Access tokens are valid for 15 minutes. The gateway re-checks device posture every 5 minutes; the last check ran at 09:58:00. At 10:00:00 an on-call engineer's role is revoked in the identity provider. The policy engine receives the revocation at 10:00:02. The engineer's token was issued at 09:58:00, and the gateway cached an allow decision for the console for that engineer at 09:59:30. What is the earliest time at which a request from the engineer to the console would be denied?
MCQ-3A zero-trust platform runs one policy engine and 200 enforcement points (sidecars, one per service). Each sidecar caches the engine's decisions for 300 s, keyed by (subject, resource, action). A contractor's access is revoked in the engine at 10:00:00. At 10:02:40 the contractor reads 8 documents through a sidecar whose cache entry for that subject and resource was written at 09:58:00. What does this show?
MCQ-11A zero-trust gateway issues a 10-hour session token after MFA. A background evaluator re-scores the session every 60 seconds: the score starts at 100, and each currently-active signal subtracts its weight. Non-compliant device −40, impossible-travel flag −30, request from an ASN unseen in 30 days −20. Policy: ≥ 70 allow, 40–69 step-up MFA, < 40 deny. At 13:00 the evaluator flagged a request from an ASN unseen in 30 days. By 14:20 that ASN is recognized and the flag has cleared, the device has flipped to non-compliant, and an impossible-travel flag has fired. What does the gateway do with the next request at 14:20?
MCQ-23An authorization service allows a request only when `analytics` appears in the subject's evaluated group list. Each decision log line records `decision`, `policy_revision`, `groups_evaluated`, and `cache_age_s`, the age of the cached group data the engine evaluated. Over a 20-minute window the service denies 3,600 requests, all at the same `policy_revision`. 3,540 of those denies show `cache_age_s` between 840 and 900 with `groups_evaluated` missing `analytics`; the remaining 60 show `cache_age_s` under 60, also with `groups_evaluated` missing `analytics`. What do the 60 fresh-cache denies establish?
MCQ-12An internal service mesh puts a policy engine in front of every request. The engine evaluates three signals and compares each signal's age to its freshness window: | signal | freshness window | | --- | --- | | device posture | 5 min | | user risk score | 15 min | | network zone | 1 min | Each resource has a tier that sets the degraded action when any required signal is older than its freshness window: | tier | required signals | degraded action | | --- | --- | --- | | 0: payments admin console | posture, risk, zone | deny | | 1: customer PII export | posture, risk | deny | | 2: internal build dashboard | posture | serve cached decision up to 60 min, then deny | | 3: public docs | none | serve cached decision indefinitely | At 09:14 the posture feed's control plane goes down; user u-4471's last posture report was at 09:02. The risk score was refreshed at 09:13 and the zone signal at 09:14:29. At 09:14:30 u-4471 requests the payments admin console (tier 0); at 09:15 u-4471 requests the internal build dashboard (tier 2). What does the engine do?
MCQ-13A payments platform's policy decision point (PDP) goes unreachable during a regional failover. The API gateway caches PDP decisions for 60 s and last refreshed its cache 45 s ago. Live traffic is 4,000 req/s to POST /transfer, which moves funds, and 12,000 req/s to GET /balance, a read-only lookup. What should the gateway do?
MCQ-16During a 90-minute outage of the company's primary identity provider, no engineer could assume the break-glass admin role: the credential vault and the role's MFA both authenticate against that same IdP, and the on-call engineer had no active session when the outage began. Which single change removes the dependency that caused the failure?
MCQ-22A payments company must show its regulator that its zero trust rollout enforces access decisions on live requests, not just on paper. Which artifact is the strongest evidence?
MCQ-21A federal agency's zero trust team finishes its CISA Zero Trust Maturity Model v2.0 self-assessment. The five pillars land at Identity: Advanced, Devices: Initial, Networks: Initial, Applications and Workloads: Traditional, Data: Traditional. The cross-cutting capabilities land at Visibility and Analytics: Initial, Automation and Orchestration: Traditional, Governance: Initial. The CIO asks for "the agency's zero trust maturity level" as a single figure for the annual report and wants CISA to certify it. What should the team produce?
MCQ-17A retailer runs 1,200 internal applications behind one VPN concentrator for 4,000 employees. The IdP already issues device-bound tokens to the 3,100 corporate-managed laptops; the other 900 are contractor-owned and cannot be enrolled in MDM. Thirty days of VPN logs show 14 web applications account for 82% of sessions. The team has 90 days and no budget for new hardware. Which first step fits these constraints?
MCQ-18A hospital's lab information system (LIS) authenticates with one static credential stored in each of 300 workstation config files. The vendor caps the LIS user table at 20 accounts, and the protocol has no session token and no MFA. The protocol does carry a free-text operator field that the LIS writes into its audit log with every result release. State auditors require the LIS's own audit log to name the technologist who released each result, and require a second factor at login. The LIS stays in service for 14 months. Which design meets both requirements?
MCQ-24A platform team runs OPA Gatekeeper as a Kubernetes admission webhook (failurePolicy: Fail, timeoutSeconds: 3, 3 replicas). A new Rego rule calls http.send against an entitlement API for every Pod CREATE. The cluster peaks at 40 Pod creates/s; the entitlement API answers in 250 ms at p99 and enforces a 20 req/s quota per client. After the rule ships, admission p99 rises from 8 ms to 2.8 s and 12% of Pod creates fail with webhook timeouts. Product requires that a revoked entitlement stop admitting Pods within 60 s. Which change removes the failures while still enforcing the rule?