Skip to content
Tech Interview Prep

Top 100 Cloud Security Engineer Practice MCQs

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.

100 questions

MCQ-1A team runs a three-tier workload on AWS: EC2 instances behind an ALB, an RDS for PostgreSQL instance, and three S3 buckets. A pen test reports three findings: the RDS instance accepts connections from 0.0.0.0/0 on port 5432, one S3 bucket allows public read, and the EC2 instances run an AMI two patch cycles behind. Which plan puts each fix on the correct side of the shared responsibility boundary?
MCQ-22A team audits all 240 security groups across three VPCs and finds no rule admitting 0.0.0.0/0 to any database port. An external scan of the account's public address space still completes a TCP handshake on 5432 against one database. That database's security group admits 5432 only from 10.0.1.0/24, the subnet of an internet-facing network load balancer that forwards to it and has no security group attached. What does this evidence establish?
MCQ-8A payments service runs as a systemd unit on an EC2 instance. Its `GET /preview?url=` endpoint fetches the supplied URL with a plain HTTP GET, sends no custom headers, does not follow redirects, and returns the first 2 KB of the response body. A penetration test called `GET /preview?url=http://169.254.169.254/latest/meta-data/iam/security-credentials/payments-node` and the response body contained the node role's `AccessKeyId`, `SecretAccessKey` and `Token`. The instance's IMDS is configured with `HttpTokens=optional`, and the node role has `s3:GetObject` on the receipts bucket. Which change makes that exact request return no credentials?
MCQ-16A bucket in a service account was made public by a Terraform change. The org's daily AWS Config rule flagged it 19 hours later, by which time a search engine had indexed it. The org runs 340 accounts, and no team has a legitimate need for a public bucket. Which change addresses the cause?
MCQ-23A platform team runs one shared interface VPC endpoint (vpce-0a1b2c) that the dev, staging and prod VPCs all route to for S3. The payments bucket policy grants s3:GetObject to Principal "*" with the condition aws:SourceVpce = vpce-0a1b2c, and account-level Block Public Access is on. A test from a compromised EC2 instance in the dev VPC downloads 40 objects from the payments bucket. Which change closes the finding without breaking the other VPCs' access to their own buckets?
MCQ-2Account 111122223333. The CI role `deploy-bot` carries one identity policy: Allow `s3:GetObject` on `arn:aws:s3:::build-artifacts/*`. Its permission boundary allows `s3:*` on `*`, and the account's SCP allows `s3:*`. The bucket policy on `build-artifacts` has one statement: Allow principal `arn:aws:iam::111122223333:role/deploy-bot`, action `s3:PutObject`, resource `arn:aws:s3:::build-artifacts/*`. A review read the identity policy, recorded the role as read-only, and signed off. An attacker using the role's leaked credentials then overwrote `build-artifacts/release.tar.gz`. What does the policy simulator return for `s3:PutObject` on `arn:aws:s3:::build-artifacts/release.tar.gz` for `deploy-bot`?
MCQ-3An AWS account holds seven IAM roles. A scanner that reads attached policies flags one privileged identity: `admin-role`, which carries `Action: *`. The escalation-relevant permissions on the rest are: | Role | Escalation-relevant permissions | | --- | --- | | `ci-runner` | `sts:AssumeRole` on `deploy` | | `deploy` | `sts:AssumeRole` on `app-role` | | `builder` | `iam:PassRole` on `app-role`; `ec2:RunInstances` | | `app-role` | `sts:AssumeRole` on `admin-role` | | `auditor` | `iam:Get*`, `iam:List*` | | `readonly` | `sts:AssumeRole` on `auditor` | Which roles can reach `admin-role`'s permissions, directly or through a chain of assume-role or PassRole edges?
MCQ-5A platform team is given iam:CreateRole in a shared AWS account so they can ship services without waiting on tickets. Security must be able to show that no role the team creates can exceed the permissions it has approved. Which requirement should the review accept?
MCQ-9Account 444455556666 owns role VendorIngest, whose trust policy is Principal: arn:aws:iam::111122223333:root with Action: sts:AssumeRole. The vendor runs a single IAM role, svc-integration, in account 111122223333 for all 400 of its customers. CloudTrail shows AssumeRole calls on VendorIngest with roleSessionName cust-8821-acme, a vendor customer you have never dealt with, succeeding and reading objects from the bucket your integration uses. Which change stops those sessions while leaving the vendor's own integration working?
MCQ-42In account 111122223333, the `thumbnailer` Lambda runs with execution role `thumbnailer-role`, which allows `s3:GetObject` on `arn:aws:s3:::uploads/*` and `s3:PutObject` on `arn:aws:s3:::thumbs/*`. The function's resource-based policy grants `lambda:InvokeFunction` to principal `arn:aws:iam::111122223333:root`. A developer's read-only role was phished; the attacker invoked `thumbnailer` with a crafted event naming an object under `thumbs/`, and CloudTrail shows the write was made by `thumbnailer-role`. Which change removes the escalation path while keeping the thumbnail pipeline working?
MCQ-25Acme runs its API in a single AWS account, with production and staging separated only by IAM policy conditions on the tag `aws:ResourceTag/Environment`. A staging deploy role's policy allows `s3:GetObject` on `arn:aws:s3:::acme-data/*` when the object tag `Environment` equals `staging`. S3 objects are never tagged, so the condition never matches and staging reads fail; to unblock a release an engineer adds a second statement granting `s3:GetObject` on the same ARN with no condition. Three weeks later CloudTrail shows the staging role read 4,100 objects from the production prefix in 9 minutes. Which change ensures a single policy mistake like this cannot reach production data?
MCQ-4The team runs a multi-tenant payments service in account 111122223333. Each pod serves one tenant and reads that tenant's KMS key ARN from a ConfigMap, but every pod runs under one shared role whose only KMS statement is: ``` {"Effect":"Allow","Action":["kms:Decrypt","kms:GenerateDataKey"],"Resource":"arn:aws:kms:us-east-1:111122223333:key/*"} ``` The account holds 12 customer-managed keys, one per tenant. A provisioning job creates a new key each time sales closes a deal and does not tag the keys. A red team compromises one pod, calls KMS directly with the pod's credentials, and decrypts all 12 tenants' data. Which change closes the finding?
MCQ-10A workload reads tenant records from a store encrypted with the provider's default at-rest key. A compromised IAM role in the same account read records for tenants it had no business reading; the storage read is in CloudTrail, but there is no key-layer event to attribute. The review requires that decryption be authorised separately from the storage read, that each decryption be attributable to a named caller, and that a record copied into another tenant's prefix must not be readable there. Which change meets all three requirements?
MCQ-41A payments service stores 40 TB across 12 million S3 objects. Each object is encrypted with a per-object AES-256 data key; the data key is wrapped by the KMS CMK `alias/payments` and stored in the object's metadata. The CMK is rotated annually and is now at backing-key version v7; every object in the bucket was written under v7. An attacker exfiltrated a full snapshot of the bucket, objects and their wrapped data keys, and the team has not finished enumerating whether the compromised role could call `kms:Decrypt` on v7. The team rotated the CMK to v8 and marked the incident closed. A dry run measured re-encryption at 250 MB/s across 8 workers. Which action is correct?
MCQ-40An SSRF in the web tier let an attacker assume the production EC2 instance role, which carries s3:DeleteObject and s3:DeleteObjectVersion on the backup bucket in the production account. In under five minutes the attacker deleted all 7 days of snapshots, created an IAM user with AdministratorAccess, and encrypted the EBS volumes. The only surviving copy was a 6-week-old pg_dump that predated two schema migrations, so recovery took 11 days. Which change bounds the next incident's data loss to the last 24 hours?
MCQ-33A payments team runs in a single AWS account. CloudTrail delivers to an S3 bucket in that account, and the workload's role holds `s3:PutObject` and `s3:DeleteObject` on it. During a breach, an attacker who had taken over an admin role called `cloudtrail:StopLogging` and then deleted 40 days of trail objects; the team learned of it 6 days later from a customer. Which change makes the trail survive a repeat of this breach?
MCQ-17An incident responder is investigating a suspected exfiltration from an S3 bucket that holds customer records. The account has a multi-region CloudTrail trail created with default settings, delivering to a log bucket; the incident window is six days ago. Querying the trail for that window returns 1,247 events: 3 PutBucketPolicy, 41 CreateAccessKey, and 0 GetObject. The role under suspicion holds s3:GetObject on the bucket. What does this evidence support?
MCQ-26A company attaches an SCP to the organization root that denies cloudtrail:StopLogging and cloudtrail:DeleteTrail. A test from a member account role returns AccessDenied for both actions. Eleven weeks later, an incident review finds CloudTrail was stopped in the management account and no SCP denial was logged. The SCP is still attached and unchanged. What should the security team do first to close the gap?
MCQ-20Repo `acme/payments-api` runs a GitHub Actions workflow with `on: [pull_request, push]` and `permissions: id-token: write`. The deploy job assumes AWS role `prod-deploy`, whose trust policy allows `StringLike` on `token.actions.githubusercontent.com:sub` = `repo:acme/payments-api:*`. A fork PR that edits `.github/workflows/deploy.yml` runs the modified workflow and its deploy step successfully assumes `prod-deploy`. Which change makes STS reject the fork PR's token while main-branch deploys still succeed?
MCQ-6A payments team runs 60 CI jobs on an EKS cluster. Each job reads a static AWS access key from a Kubernetes secret, and a script rotates those keys every 90 days. CloudTrail shows one key, created 26 months ago and absent from the rotation script's inventory, calling S3 GetObject from 203.0.113.44, an address outside the VPC, with the same user-agent string the CI runner sends. The cluster already has an IAM OIDC provider registered. What should the team do first?
MCQ-21A monitoring vendor connects to customer AWS accounts by assuming a role in each one. It asks every customer to trust the same vendor account ID, 111122223333, and to set an sts:ExternalId it generates per customer. Your change request creates the role in all 42 accounts of your organization, attaches the AWS managed ReadOnlyAccess policy, and sets the trust policy principal to 111122223333 with no sts:ExternalId condition. The integration only reads CloudWatch metrics and tags from 3 accounts. Six months later, another customer of the vendor, who has your role ARN from a public post, asks the vendor's platform to connect to your account. What happens?
MCQ-24A workload in the build account was compromised and pushed 400 GB to an external object-storage endpoint over six days. Ingress rules were tight; egress was unrestricted and no destination was recorded. The workload's legitimate outbound traffic goes to internal services and to one package registry that resolves to addresses in a shared CDN range. Which change would have stopped the transfer?
MCQ-13A red-team exercise compromised the CI runner that builds `registry.internal/checkout`. Using the runner's own registry credentials, the team pushed a modified image to `checkout:stable` at 02:47; the push came from the runner's IP with the runner's token, so the registry audit log shows it as a normal build. The deployment controller resolved the tag at 03:00 and the pod ran for 9 minutes. The admission controller is deployed in audit mode, it logs policy violations but does not block, and no build step produces a provenance attestation. The modified image is the legitimate base plus a backdoor binary, so it introduces no new CVEs. Which change to the deployment path would have stopped this pod from running?
MCQ-12A pre-receive hook on the shared origin scans the tip commit of each pushed ref for credentials. Last quarter it blocked 47 pushes, but 9 credentials still reached the remote. In all 9, the secret was added in an intermediate commit of a multi-commit push and deleted in a later commit of that same push; pushes average 6 commits. The team's rule is that no credential may reach the shared origin. Which change closes the gap?
MCQ-11A Node.js 20 service on Kubernetes reads its Postgres password from an environment variable injected with envFrom: secretRef. An incident review confirms three places the current value leaks: the error-tracking SDK's crash payload includes process.env, any process running as the app user inside the container can read it from /proc/self/environ, and the app's startup log prints the parsed config object with the password in it. Which change removes all three leaks?
MCQ-14A 24-node Kubernetes cluster runs the company's own API pods and third-party tenant batch jobs on the same nodes. The API pods have a 40 ms p99 budget with no headroom, so they must keep running directly on the host kernel with no syscall interception or virtualization. The threat model treats tenant code as untrusted, and the team wants a boundary that holds even when the next kernel privilege-escalation CVE lands. Which change gives tenant code that boundary without touching the API pods' runtime?
MCQ-36A team's CI builds `FROM node:20@sha256:9f2c…`, installs npm dependencies with `npm ci` against a committed `package-lock.json`, and provisions networking with a public Terraform module pinned `version = "~> 5.0"`. The module maintainer publishes v5.4.2 with a provisioner that exfiltrates the CI job's environment. The team changes nothing in the repository. Which dependency paths can pick up the new publish on the next pipeline run?
MCQ-70A CI system runs pull-request builds from forks on 12 self-hosted runners. Every runner shares one long-lived IAM role that can deploy to production, and runner egress is unrestricted. A dependency's post-install script read the role's credentials from the instance metadata endpoint and POSTed them to an external host; 40 minutes later CloudTrail shows that role deploying from an unfamiliar IP. Which change removes the exposure that made this possible?
MCQ-71A platform team keeps the production RDS master password in a `random_password` resource inside the network configuration's Terraform state, stored in an S3 bucket whose policy grants `s3:GetObject` to 3 IAM roles. The application configuration reads that password through a `terraform_remote_state` data source pointed at the same bucket, and the application's own state lives in a second S3 bucket whose policy grants `s3:GetObject` to 9 IAM roles. Two of those 9 roles also appear in the network bucket's policy. Both buckets use SSE-S3. Before the application configuration was added, only the network state held the password. How many distinct IAM roles can read the plaintext password from a state file after the change?
MCQ-80A deploy access key turned up in a public fork. It had been passed into a Terraform provider block, so it sits in 74 retained versions of the state file in a versioned S3 bucket that roughly 60 engineers can read; `sensitive = true` redacted only the plan output. Nine consumers used the key and two of them were undocumented. Which change addresses the cause?
MCQ-77A database credential is read by 22 services across three accounts. Last quarter's rotation took 7 hours because every consumer had to be redeployed, and during the incident review nobody could say which service had read the credential. Which statement should the review accept?
MCQ-100The payments service runs on 40 EC2 instances behind an ALB and opens about 2,000 new database connections per second. Its database password is a constant in the source tree. The team is moving it to the managed secret store, which supports rotation with an SNS notification, revokes the previous version the moment rotation completes, and allows 100 reads per second per account. Rotation must reach every instance within 60 seconds with no failed request and no deploy; the pipeline is change-gated and takes about 12 minutes. Which option holds up?
MCQ-84A document service authorizes writes in three steps: it calls GetObjectTagging to read the object's owner tag, compares that tag to the caller's tenant ID, then calls PutObject. A pentest measured the gap between the tag read and the write at p50 12 ms and p99 240 ms under load; the separate tagging endpoint, owned by another team, can reassign the owner tag in about 5 ms. Racing the two, the pentest got a write accepted under a tag that had already been reassigned to a different tenant. The tagging endpoint cannot be changed. Which change closes the race?
MCQ-83A shared logging service holds a role with s3:PutObject on every customer bucket. It authenticates each caller, then writes to the bucket name given in the request. A customer's CI job passed another customer's bucket name and the write succeeded. Bucket names are chosen by customers and carry no account prefix, and every caller is a legitimate customer of the platform. Which change stops the cross-customer write?
MCQ-88A team runs an admission webhook that verifies cosign signatures on images. The webhook's `rules` list only `Deployment` in `apps/v1`; it is set to `failurePolicy: Fail` and covers all namespaces. Last month a developer created a `CronJob` that pulled `docker.io/library/redis:7`, an unsigned public image; it ran for 9 hours before a manual audit caught it. Which change addresses the cause?
MCQ-87A payments team's CI resolves dependencies through a virtual repository that aggregates the public npm registry and the team's internal registry, selecting the highest version across both. The internal registry is a standalone repository, not a proxy of npmjs. The build manifest declares `@acme/ledger-client: ^2.0.0`; the internal copy is 2.4.1. An attacker registered the `@acme` scope on npmjs and published 9.0.0. After the incident, the team changed the CI repository's ordering rule to prefer the internal registry when both sources publish the same version. The next build, on a fresh runner with an empty package cache, still resolved to the attacker's 9.0.0. Why?
MCQ-91A cloud estate has 214 open scanner findings. The team's graph tool models identities, resources and network reachability as nodes and edges. Six findings touch customers-prod, the RDS instance holding the customer records, and the graph links them like this: - F-31 (critical, CVSS 9.1): the S3 bucket archive-exports serves historical exports to anonymous callers. No role, resource or network path in the estate trusts that bucket. - F-54 (medium): the public HTTPS load balancer alb-web forwards to the web instance group and sits behind no authenticating layer. - F-57 (medium): the public API Gateway exposes an unauthenticated route to the reporting Lambda. - F-52 (medium): the web instance group's instance profile can sts:AssumeRole on ops-shared. - F-58 (medium): the reporting Lambda's execution role can sts:AssumeRole on ops-shared. - F-53 (medium): ops-shared carries rds-db:connect on customers-prod and can read the customers-prod credential secret. These are all the edges into ops-shared and customers-prod. The team can ship exactly one change this week, and it has to leave no externally reachable path to customers-prod. Which change does that?
MCQ-90An intrusion was traced to the integrity controls on an object-storage archive: records were modified by an identity that had legitimate write access to the bucket, and the storage service's own checksums were recomputed along with the objects, so the change left no trace. Which remediation would have detected the modification?
MCQ-92A remediation service consumes findings from a queue that delivers at-least-once, and the team must settle its design before the service handles regulated data. Actions include revoking IAM policies and deleting S3 buckets. The scanner re-emits every open finding hourly, the queue's deduplication window is 5 minutes, and the pipeline emits a few thousand findings a day. Last month a duplicate delivery re-ran a revoke and removed a policy that had been re-added between the two runs. Which design should the team adopt?
MCQ-43An order service publishes `order.created` events to a shared SQS queue (default 4-day retention). The consumer reads `tenant_id` from the message body and writes the order into that tenant's schema. Four IAM principals hold `sqs:SendMessage` on the queue: the order service, a CI/CD deploy role, a nightly data-export job, and a legacy billing bridge. Only the order service produces `order.created`; the other three grants were added during a migration and never removed. A pen test published a well-formed `order.created` as the billing bridge with another tenant's `tenant_id`, and the consumer wrote into that tenant's schema. Which change removes the ability to do that?
MCQ-81A page is served with this header: ``` Content-Security-Policy: script-src 'nonce-8f3a91' https://cdn.example.com; object-src 'none' ``` The nonce is a constant baked into the page template, so every response carries the same value. `cdn.example.com` is a CDN that serves user-uploaded files under `/u/`. A stored-XSS flaw lets an attacker place markup in a comment field, which is rendered into the page body. The attacker submits three comments: 1. `<script>fetch('/steal')</script>` 2. `<script src="https://cdn.example.com/u/9f2.js"></script>` 3. `<script nonce="8f3a91">fetch('/steal')</script>` How many of the three execute in a victim's browser?
MCQ-86A Node 18 service writes one pino-http line per request with `method`, `url`, `status` and `durationMs`. Password-reset links look like `https://app.example.com/reset?token=<JWT>`, and a 24-hour scan of 90,000 access-log lines found 3,412 carrying a live reset token in the `url` field. The redact config is `["password", "token", "authorization", "cookie"]`. The SLO dashboard and incident forensics need the request path on every line, but no query string may be logged. Which change removes the tokens and still leaves the path on every line?
MCQ-85A payments API calls a central authorization service on every request. The gateway is configured with `on_authz_unavailable: serve_cached`, a decision-cache TTL of 1800 s, and `on_cache_miss: deny`. During a 20-minute authorization-service outage the gateway handled 1,900 requests: 1,842 were allowed from cache and 58 were denied. The change note states that the gateway 'fails closed'. What should the review conclude about the change note?
MCQ-51An invoices API verifies the JWT at the gateway and logs the `sub` claim; the pen-test token carried a valid `invoices:read` scope. Changing `/invoices/1042` to `/invoices/1043` returned another tenant's invoice with HTTP 200. The handler runs `SELECT * FROM invoices WHERE id = ?`, and 340 handlers across the service follow the same shape. Which change closes the finding?
MCQ-52A regulated workload runs behind a legacy service whose only outbound path is a nightly batch job, and the identity store is reachable only from the auth service. An incident review found an access token copied from a log stayed valid for 19 hours after the account was disabled. The team must bring the post-disable window under 15 minutes without changing the legacy service. Which token lifetime and refresh design does that?
MCQ-53An internal gateway authenticates once to the auth server and receives a token carrying the scopes for all 14 internal APIs it can call. It forwards that same token, unchanged, in the Authorization header to every downstream service. One downstream service writes full request headers, including that header, to a log store every engineer can read. Which change actually reduces what an attacker who reads that log can reach?
MCQ-55A payments platform runs 40 services on Kubernetes. Mutual TLS is issued by cert-manager: each service presents a 24-hour certificate bound to its SPIFFE identity, and the payments service authorizes callers by matching that identity against an allowlist. A namespace rename changed the frontend's identity from spiffe://cluster.local/ns/frontend/sa/web to spiffe://cluster.local/ns/web/sa/web, and cert-manager reissued its certificate under the new identity. Checkout now fails: every frontend call reaches the payments service, completes the TLS handshake, and is rejected with HTTP 403. Which diagnosis fits?
MCQ-56A checkout service is protected by an edge WAF rule set and an in-process input validator. One Helm chart renders `values.yaml`'s `blockedPatterns` list into both the WAF config and the service's ConfigMap, in the same deploy. A merge widened `blockedPatterns`; the next deploy pushed it to both, and a SQL-injection payload that both had blocked in staging reached the database in production. Which change makes the two layers independent?
MCQ-57An engineering org declared its zero-trust migration complete after putting the customer-facing web application behind an access broker that checks identity and device posture on every request. The broker's own logs for the trailing 30 days cover 2,000,000 access decisions to production: | Path | How access is granted | Decisions | | --- | --- | --- | | access broker | identity and device posture checked per request | 1,100,000 | | corporate VPN | MFA at login, then flat access to the internal production subnets for the rest of the session | 800,000 | | build system | static bearer token over a site-to-site tunnel, no posture evaluation | 100,000 | The CISO set the program's exit criterion: no production access is granted on the basis of network location, or on a credential not bound to a verified device. The team's coverage number is the share of these decisions made with full identity and device context. Which statement about the program's status is correct?
MCQ-54A payments platform serves TLS from 247 certificates that a quarterly network scan finds; the CMDB lists 210, each with automated renewal and a 30-day expiry alert. Of the 37 the scan finds and the CMDB does not, 9 are on cloud load balancers renewed automatically by the provider's ACME client, and 28 are hand-issued on appliances and brokers with no renewal automation and no owner. Last month one of the 28 expired on a Saturday and caused a 5-hour outage. The team proposes to import all 37 into the CMDB and set a 30-day calendar reminder for each. Which assessment of that proposal is correct?
MCQ-76A three-tier workload (web, app, data) runs in one VPC and will hold regulated data next quarter. The team must commit to one network boundary design now, and the threat model assumes an attacker who has already compromised a host in the app tier. Which design is defensible?
MCQ-75A security group is evaluating its 'deny' control for non-compliant business requests before relying on it. Last quarter three requests were denied outright; each requester then stood the workload up in a personal cloud account, and one held customer data for seven months before a billing query surfaced it. The underlying business need in each case was legitimate: only the proposed implementation broke the standard. Which design for the deny control holds up?
MCQ-98A multi-account AWS organisation has recorded four accidental public exposures of S3 buckets in eighteen months, each one caused by a team setting a bucket ACL or bucket policy by hand under pressure during an incident. Which control set most reliably prevents the fifth exposure and still surfaces whatever slips past it?
MCQ-7A review of a 200-bucket AWS account found that a bucket created six months earlier by a provisioning automation was publicly readable and held 40,000 customer invoices; an external researcher reported it before the nightly Config rule did. Three older buckets had been public since 2021, about a third of the account's buckets are created by application teams with their own Terraform, and account admins have twice disabled controls they found inconvenient. Which control should the team adopt?
MCQ-18A red-team exercise against a production AWS account: the attacker used a leaked access key and reached administrator 20 minutes later. The account's detection pipeline measured 8 minutes for CloudTrail log delivery to S3, a correlation rule that evaluates on a 5-minute batch, and a 22-minute median on-call acknowledgement. The team wants one change that brings the pipeline from action to a human acting on it inside the 20-minute window. Which change does that?
MCQ-34A managed cloud-log detection service watches 60 AWS accounts with its default rule set, and one on-call engineer owns the page queue. A 90-day review shows the volume-anomaly rule fired 1,140 times with 6 alerts actioned, the credential-used-in-a-new-region rule fired 40 times with 18 actioned, and the audit-logging-disabled rule fired 3 times with 3 actioned. The provider ships the volume rule at HIGH severity and the other two at MEDIUM. The one confirmed intrusion in the window, permission enumeration from a new region, was paged at 04:00 and sat 9 hours behind a 42-page overnight queue. Which change should the team make first?
MCQ-48A post-incident review finds that an intrusion used a technique from the team's mapped set of 40, and that a detection rule for that technique exists and passes its unit test against a synthetic sample event. No alert fired for the account involved, and the review cannot tell whether the event was never collected, the rule failed to match, or the alert lacked the context to act on. Which change addresses the cause?
MCQ-27A 3,000-person company federates its AWS estate to a single Okta tenant. At 02:14 an attacker holding an admin session in Okta (stolen cookie, MFA already satisfied) adds a SAML claim mapping that grants the group `contractors` the production `OrganizationAccountAccessRole`. At 02:31 CloudTrail records `AssumeRoleWithSAML` for `d.reyes` with MFA=true and source IP 203.0.113.44, the corporate VPN egress, followed by `CreateUser` calls. The Okta system log records the 02:14 change with actor `admin@corp`. The SOC's rules cover impossible travel, MFA failures, root-account use and IAM policy changes in AWS. Reviewing the next morning, the SOC finds that no alert fired. Which explanation is correct?
MCQ-28An administrator's cloud console session was taken over. The attacker ran a reverse proxy in front of the tenant's real login page; the administrator typed their password into it, the proxy relayed the push prompt to their phone, and the administrator approved it after the proxy displayed the correct two-digit number. The identity provider logged the sign-in as MFA-satisfied and device-compliant, because the proxy forwarded the session from the administrator's managed, posture-passing laptop. The password was unique to this account and no malware was found on the laptop. The tenant has 12 administrators, who work from home two days a week, so there is no fixed corporate egress IP. Which change removes the path the attacker used?
MCQ-29A payments account holds cardholder data. Last quarter the corporate identity provider was down for 3 hours and could not be repaired from the cloud console; the account's break-glass credential was unusable because its hardware factor had expired, so nobody could reach the account for the whole outage. The team proposes a change to break-glass access. Which statement should the review accept?
MCQ-15A production AWS organisation spans 60 accounts. During a sev-1 the on-call responder must hold administrator access in the affected account within two minutes. For every use the auditor must be able to name the engineer, the reason, and the time the access ended. The identity provider failed during the last sev-1, so the path must not depend on it; a session in one account must not reach the other 59; and the last two sev-1s ended with the break-glass session still active the next morning because nobody closed it. Which control set meets all of these?
MCQ-31A platform team of 30 engineers replaces standing AdministratorAccess with just-in-time elevation: an engineer self-serves a 4-hour grant through an internal broker, the grant is logged with a stated reason, and it expires automatically. Three months later the quarterly review runs a query for identities that can obtain administrative permissions and gets back 30: the same number as before the rollout. The broker's automation role carries AdministratorAccess, its trust policy names the engineers' identity group as a principal, and it sets no session-duration or MFA condition. Which reading of the result is right, and which change removes the standing exposure?
MCQ-30An audit of 50 engineers with three or more years of tenure at a 1,400-person company found 640 standing cloud grants. 210 map to the engineer's current team through an Okta group-to-cloud-role mapping. The other 430 are direct IAM policy attachments created by a legacy onboarding script at hire; 380 of them grant access to a team the engineer has since left. Changing an engineer's team in Okta updates the group-derived grants within five minutes and leaves the direct attachments unchanged. HR deactivation removes the group-derived grants within five minutes and leaves the direct attachments in place. The IdP's access-review campaign enumerates group memberships only. Which change removes the largest share of the unjustified access?
MCQ-32Six months after a destructive change, the cloud audit log shows the call came from the pipeline's deployment role and nothing else. The CI system keeps its logs for 90 days; the cloud audit log is kept for 400. The team wants the next investigation to name the person from the cloud audit entry alone. Which change achieves that?
MCQ-35At 02:14 GuardDuty reports that credentials for the instance role `ci-deploy` are being used from 203.0.113.77 to call DescribeInstances, and the instance's CPU has been pinned at 100% for 40 minutes. The role's MaxSessionDuration is 12 hours, and the session the attacker is using was issued at 01:34 with that maximum. The instance's EBS volumes are flagged delete-on-termination. The on-call engineer's runbook says "terminate the instance, then investigate." What does terminating the instance at 02:20 actually accomplish?
MCQ-19At 09:00 an attacker copies a long-lived IAM user access key from a compromised laptop. At 09:12 the attacker calls sts:AssumeRole on the role prod-app-role, whose MaxSessionDuration is 12 hours, and receives session credentials. At 09:30 the team deletes the access key and confirms it no longer authenticates, but at 10:00 the attacker's session is still calling the API. The role is used by a production service that must keep running. Which action stops the attacker's session while leaving that service working?
MCQ-78An IAM access key for a CI service account was committed to a public repo at 09:14 UTC and deleted at 09:20. CloudTrail for that key ID shows its last signed request at 09:18 and an AssumeRole call at 09:16 that returned credentials with a 12-hour duration. At 09:55, from the same source IP as the 09:16 AssumeRole, the trail records iam:CreateUser followed by iam:CreateAccessKey. The ticket was closed at 09:26 with the note 'key deleted, exposure window 6 minutes.' At 11:00, which statement about what is still live is correct?
MCQ-79A cloud security engineer must justify how the blast radius of a leaked access key is handled. The key belongs to a CI identity, was committed to a public repository, and stayed there for 19 days. The identity can assume a deploy role in the data account through a trust policy with no external ID and no source-account condition, and that account's S3 data events are turned off. `GetAccessKeyLastUsed` returns no date, and the credential report cannot be regenerated for another three hours. Which justification is sound?
MCQ-66A long-term cloud access key is found in the commit history of a public repository about twenty minutes after it was pushed. What is the correct remediation?
MCQ-94An intrusion is traced to management ports open to 0.0.0.0/0. What is the correct remediation?
MCQ-96An EC2 instance carrying an instance profile is sending outbound crypto-mining traffic and is being treated as compromised. Which containment sequence stops the blast radius without destroying the evidence the investigation needs?
MCQ-95A platform team's secret-access process requires a ticket with a two-day turnaround. Telemetry shows 90% of engineers copy credentials into a shared document instead, which takes about two minutes. The security lead asks what to do about the finding. Which response is sound?
MCQ-93During a credential-compromise incident, an on-call engineer spent 40 minutes trying to isolate the affected service account because the runbook said "isolate the affected account" but named neither the CLI command nor the approver. The proposed replacement has numbered steps, exact commands with expected output, decision points, and an approval path. Before merging, the team runs a drill: an engineer who did not write the new runbook completes 12 of 12 steps in 6 minutes; the same engineer using the old runbook completed 4 of 12 steps in 40 minutes. Which review decision is correct?
MCQ-64A payments team runs a 90-minute quarterly tabletop. In the last one, the on-call engineer described failing over the primary Postgres database to the replica, and the report said "runbook reviewed, no gaps." Three weeks later a real primary failure took 47 minutes to fail over: the on-call's IAM role could not promote the replica, and the runbook's promotion command pointed at a host decommissioned two quarters earlier. The post-incident review asks what change to the exercise would have surfaced both gaps before the outage. Which change should they make?
MCQ-63An incident review found that the compromised instance was replaced by autoscaling on day 2, the investigation began on day 5, and the account's 7-day control-plane log retention covered only the first half of the intrusion. The team has one forensic-readiness budget line and must justify how to spend it. Which justification is sound?
MCQ-62A payments company classifies incidents by observable impact. Sev1: confirmed or suspected unauthorized access to customer data. Sev2: degradation of a customer-facing service with no data exposure. Sev3: internal-only impact, or a suspected misconfiguration with no evidence of data access. Sev4: cosmetic or planned maintenance. Any responder may raise severity; only the incident commander may lower it. At 02:00 an alert shows a CI runner issuing GETs against the customer-exports bucket, which holds 40,000 customer records. The on-call engineer labels it Sev3, 'suspected misconfigured batch job', and does not page the security team. At 08:00 forensics confirms the records were read. Which severity was correct at 02:00?
MCQ-37A platform team runs a plan-time policy scan in CI before every `terraform apply`. Over 18 months the pipeline applied 380 resources. A separate runtime posture scan now reports 57 resources in the same cloud accounts that have no matching Terraform state, and 9 of those 57 allow public ingress. The CI gate has passed every plan. What should the review conclude?
MCQ-60The team runs production in AWS account 1111 and DR in account 2222. Both hold the same 2.4 TB of cardholder documents; 2222 receives them by S3 cross-region replication. Production has 6 SSO roles with MFA, quarterly access reviews, GuardDuty and Security Hub enabled, and no open findings. Account 2222 was built in 2021 from a copy of the production template and has drifted: 31 IAM users, 9 with console access, 4 of those without MFA, GuardDuty off, last access review in 2021. A scan of 2222 also shows the replica bucket policy grants s3:GetObject to arn:aws:iam::4444:root; account 4444 was a partner decommissioned 14 months ago and its root credentials are unmanaged. The team has one engineer and two weeks before the workload handles live cardholder data. Which change should they make first?
MCQ-59A platform team has one engineer-week before an audit. Their estate (hypothetical): 61 Lambda functions on a runtime with 5 months of support left, 14 of which cannot be updated in place because the provider stopped allowing updates to that runtime; 12 RDS instances on a major version with 3 months of support left; and 210 container images based on a base image with 18 months of support left. The provider patches the managed service control plane and engine. Which control should they build first?
MCQ-61A payments team wants to run a query-plan regression suite against a nightly copy of the production Postgres cluster: 2.1 TB holding card tokens and customer PII: in a staging account. The suite only reproduces the production plan regressions when it runs against the real data distribution and referential skew. The copy lands in staging first and a masking job runs there afterwards; that job failed silently for eight months last year, exiting 0 while leaving identifiers in place. The staging account grants read access to 300 engineers through one shared role. Which change should the review require before approving the pipeline?
MCQ-74A multi-tenant service that stores cardholder data for 1,200 enterprise customers launched six weeks ago with API audit logging, a passing external penetration test, and a clean compliance scan of the production account. A service account was then compromised and read customer records for three weeks before it was disabled. The audit log shows the account's creation and permission changes, but the team cannot identify which of the 1,200 customers' records were read. Which control, had it been required at launch, would have answered that question?
MCQ-49A payments team must keep EU cardholder data inside the EU. A review of their managed Kubernetes service found the cluster, node pools, and the managed Postgres primary and its backups all in eu-central-1, but the service's audit-log stream is delivered to a default sink in us-east-1, and the vendor's support console shows engineers outside the EU can read cluster metadata and node logs. Which action should the team take?
MCQ-50A production cloud environment needs a policy for phishing-resistant MFA on privileged administrative access. Which policy is correct?
MCQ-67A fintech holds SOC 2 Type II. The scope statement names a single AWS account, prod-1111. A data map shows customer PII in three stores: the prod-1111 RDS instance, a Snowflake warehouse loaded by nightly exports, and a support team's S3 bucket in account corp-2222. Last quarter an attacker with a leaked CI deploy key read the Snowflake export bucket. A prospect's security review asks whether the certification means its customer data is protected. Which response is correct?
MCQ-68A 40-engineer platform team is onboarding a SaaS log-analytics vendor that will process regulated data. The vendor's published tier matrix is below. The team's identity provider requires SAML for every production system, and the quarterly access review requires an exportable audit log. The contract on the table is the Standard tier. Which action is defensible? | Capability | Standard | Enterprise | | --- | --- | --- | | SSO/SAML federation | no | yes | | audit-log export | no | yes | | deletion on request | yes | yes | | sub-processor list | public | public |
MCQ-69An 18-month insider review at a payments platform covers 23 incidents. Fifteen were accidental: engineers who had moved to other teams months earlier still held production admin and ran a destructive script against the wrong cluster. Six came from leavers whose production admin access was live 30 to 90 days after their last day. Two were malicious, both surfaced by behavioural monitoring added six months ago, which has since raised 4,100 alerts. Forty engineers hold standing production admin, and the platform ships about 200 production deploys a day. Which change should the review recommend first?
MCQ-72A company runs 60 AWS accounts. An hourly job compares each account's spend with its trailing 7-day average and pages the finance on-call when spend exceeds 3x that average; accounts with no spend in the window are skipped because the average is undefined. A dormant account (no workloads, no spend for 90 days) is compromised through a leaked deploy key, and a miner spends $700/day for 12 days before a security engineer spots the account in a cost review. The finance on-call triages cost pages during business hours and escalates to security only when asked. Which change would have produced the earliest signal that security acts on?
MCQ-73A quarterly review of a 44-account AWS organization reconciles the account inventory against the ownership registry. The registry requires every account to name an owner who re-attests every 90 days: 31 accounts have an owner who responded, and 13 have no owner responding. The posture scanner is deployed to 40 accounts and reports every one of them as passing; the 4 accounts it does not cover are all among the 13 with no owner responding. One of those 4 is a project account from a cancelled initiative: it still holds a cross-account role that the production account trusts and a snapshot of the customer database, and its last owner left 2 years ago. Which action does this evidence justify?
MCQ-82An access review at a payments platform (400 engineers, role-based authorisation only) counted 240 roles: 180 are held by exactly one person, the other 60 by whole teams and matching a job function. The finding that triggered the review was a payments engineer's break-glass grant, created 11 months earlier for a one-week backfill, still working, with the same grant carrying read access to the customer PII table. The quarterly review had signed off the role list twice since the grant was made. Which change addresses the cause of the finding?
MCQ-65A 90-item security debt register is being reviewed ahead of a re-org. 22 items have a named owner and 14 have a review date. Across the last 18 months, post-incident reviews linked 6 production incidents to items already on the register: all 6 were rated medium when raised, all 6 had been open more than 12 months, all 6 had a named owner, and none had a review date. The write-ups trace initial access to a build-system deploy credential and an idle non-production account. The quarterly compliance scan, which covers internet-facing production only, reports no critical findings. Which action does this data support?
MCQ-58A team ranks its 4,000 open findings by CVSS base score and works the top 20 each sprint. Last quarter an attacker pivoted from a compromised build server to a CVSS 5.4 flaw in an internal-only service that held no regulated data. The flaw had no public exploit, was 14 months old (340th by age), and sat at position 1,204 under the current ranking. The backlog holds 380 internet-facing findings, 22 KEV-listed findings, and 1,900 internal-only findings with CVSS ≥ 7.0; the build server's subnet has 12 open findings across 5 hosts. Which change would have moved the exploited flaw into the top 20?
MCQ-89A payments service runs a nightly SCA scan. This week it reports 512 findings. Call-graph reachability analysis marks 18 as reachable from the service's entry points: 11 only from an admin endpoint behind the corporate VPN that requires an authenticated internal role, and 7 from the unauthenticated public checkout path. The 11 admin-path findings carry CVSS 8.1–9.8; the 7 public-path findings carry CVSS 6.5–7.5. The team ships one dependency upgrade per day. Which plan matches the exposure?
MCQ-47A cloud security team reports quarterly to its CISO. Across four quarters, findings closed rose from 220 to 410, median time to remediate a critical finding rose from 6 to 19 days, accounts with data-plane logging enabled went from 31 to 34 of 96, and standing production roles with administrator policy held at 12. The CISO asks what next quarter's report should lead with.
MCQ-45A platform team's security reviews all land at implementation-complete. Last quarter, findings at that stage cost a median three weeks of rework, one authorisation-model change took six weeks, and three product teams skipped booking a review rather than absorb the delay. The security group has two engineers covering forty teams. Which change should the security lead make?
MCQ-46A cloud team's risk register holds 140 accepted risks, none with an expiry or review date. An audit finds that 31 name systems decommissioned over a year ago, 4 name controls that have since become mandatory company-wide, and 105 still describe live systems whose stated conditions hold. The CISO wants the register cleaned up. Which action is correct?
MCQ-39A payments provider inventories 3,000 data stores, all currently on one control profile: a quarterly access review, 7-year retention, and a provider-managed key. Of those stores, 2,400 hold public marketing assets, 560 hold internal telemetry, and 40 hold customer identity documents. The security team tags each store with its class and proposes driving review cadence from the tag for this cycle: annual for public, every 6 months for internal, quarterly for restricted, with retention and key policy left unchanged. How many access reviews a year does the tiered cadence produce?
MCQ-44A payments team threat-models its AWS workload before it starts handling cardholder data. The first pass draws VPCs, subnets and security groups. A red-team exercise then reads the cardholder-data S3 bucket using the CI deploy role's credentials, without crossing any boundary the model drew. Which change to the model would have surfaced that path?
MCQ-38An 8-person team runs a multi-tenant SaaS on one Postgres 16 database: shared schema, a tenant_id column on every table, 4,000 tenants, and a single connection pool whose role owns those tables. Last week a support engineer's ad-hoc export script omitted the tenant predicate and returned 12,000 rows spanning 300 tenants; all 340 query sites in the application had the predicate. The team wants the next such script to be unable to return another tenant's rows, without touching the 340 query sites and without per-tenant migrations. Which change achieves that?
MCQ-99A payments company's security team has sent leadership the same quarterly report for three quarters: 4,000 open findings ranked by CVSS severity, 37 of them critical. No remediation budget has been approved. Asked why, the CFO said she cannot tell which of the 37 would actually cost the company money. A year earlier, a one-page brief describing a single path, a contractor's still-valid credential reaching the cardholder database, was funded within a week. The team has since confirmed a second live path: an internet-facing VPN missing a patch that would let an attacker reach the same database. Which change to the next report is most likely to get a funding decision?
MCQ-97A payments platform's design review has $100,000/year to spend on access controls and two proposals in front of it. (A) Mandatory 90-day password rotation for all 300 staff: 200 engineer-hours at $120/hour plus 340 measured helpdesk reset hours at $60/hour. The platform already enforces phishing-resistant MFA, so rotation has prevented no incident the team has seen. (B) Hardware security keys for the 15 engineers with production deploy rights: 15 keys at $55 each, 40 rollout hours at $120/hour, and $2,000/year for replacements. A stolen deploy credential caused one incident in the last three years, costing $210,000. Which should the review fund?