Skip to content
Tech Interview Prep
Technical interview guide

Cloud Security Posture Management (CSPM)

Continuously scanning cloud environments for misconfigurations before they're exploited.

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

Scope: AWS Security Hub CSPM, Microsoft Defender for Cloud, Google Security Command Center, and NIST SP 800-137 guidance current 2026-09-01.

Curated: · Written: · Reviewed:

25 questions

MCQ-1A team runs a CSPM tool that calls read-only cloud APIs on a 6-hour schedule (00:00, 06:00, 12:00, 18:00) and evaluates each resource's configuration against a control set that includes a public-read rule for S3 buckets. On Tuesday a bucket's policy was changed to allow public read at 09:00, 4 GB of objects were downloaded at 09:20, and the policy was reverted at 09:40. The 12:00 scan reports the bucket as compliant and raises no alert. Which explanation fits the evidence?
MCQ-21A team's CSPM tool runs a full posture assessment every 6 hours (at 00:00, 06:00, 12:00 and 18:00 UTC) and re-evaluates a resource's findings only during those runs. At 09:40 an engineer removes the public ACL from a storage bucket that the 06:00 run flagged as publicly readable. At 10:15 the dashboard still lists the finding as open and the on-call engineer asks what to do. What is the correct read of the situation?
MCQ-7A team marks a Security Hub finding for a public S3 bucket as SUPPRESSED at 09:00 on Monday, and it drops out of their WorkflowStatus = NEW filter. The bucket is not modified, and the account is not part of any cross-Region aggregation. At 21:00 the same finding ID is back in that filter. Security Hub evaluates this control on its 12-hour schedule. What explains the reappearance?
MCQ-11A CSPM tool writes findings to a table with columns severity, status, owner, created_date, due_date. Until now, owner and due_date were left null, and the nightly escalation job, which runs `SELECT * FROM findings WHERE due_date < CURRENT_DATE AND status = 'open'` and notifies the owner, has never sent a notification. The team enables a policy that backfills due_date as created_date + 7 days for high, +30 for medium, +90 for low, and assigns owners from resource tags. On the first night after the backfill, 2,400 findings have a due_date earlier than today. What does the escalation job do?
MCQ-3A delegated Security Hub administrator in us-east-1 enables central configuration and configures the Production OU to disable a noisy control. The team links us-east-1, us-west-2, and eu-west-1 to the home Region. A week later, the control still generates findings in eu-central-1 for Production OU accounts, but not in us-west-2 or eu-west-1. What explains the eu-central-1 behavior?
MCQ-16A team tracks Microsoft Defender for Cloud secure score for a subscription with 200 in-scope resources. After they mark 60 resources as exempt from recommendations they cannot fix, the score moves from 55% to 71% with no other change. What does that movement mean?
MCQ-17A subscription's secure score for the control 'Storage accounts should use a customer-managed key' reads 40% on Monday and 50% on Tuesday. On Monday, 80 of the 200 storage accounts assessed by that control were healthy. Between the two readings no storage account was created, deleted or reconfigured, and no subscription was added to or removed from the connector. Which change accounts for the rise?
MCQ-8A cloud security team triages findings from one AWS account. One record reads: role `report-gen` holds `s3:GetObject` on 412 buckets, and CloudTrail data events show it read 3 of them in the last 90 days. The team assigns each record to the tool whose assessment surface produced it. Which assignment is correct?
MCQ-9Two findings are in the triage queue. **Finding A**: image `worker:4.2`, reported as `PyYAML 5.3.1` (CVE-2020-14343). `pip list` inside the running container does not list PyYAML; the string `PyYAML==5.3.1` appears only in `/app/requirements.lock`, which the Dockerfile copies in for audit. The service owner says the lockfile will not be removed this quarter. **Finding B**: image `api:1.9`, reported as `libssl1.1 1.1.1k` (CVE-2022-0778). `dpkg -l` inside the running container lists `libssl1.1`, and `ldd /app/api` shows the binary links `libcrypto.so.1.1`. The service owner will not refresh the base image until Q3. Which classification is correct?
MCQ-2A CSPM tool scans 9 AWS accounts and reported 1,240 findings on Monday, 380 of them S3 findings derived from bucket policies and ACLs. On Tuesday the platform team attached an SCP at the organization root that denies s3:GetBucketPolicy and s3:GetBucketAcl to every principal except the break-glass role; no other configuration changed. On Wednesday the tool reports 860 findings, no errors, and a successful scan timestamp. Using the break-glass role, an engineer confirms the S3 bucket policies are byte-for-byte identical to Monday's. What most likely explains the drop from 1,240 to 860?
MCQ-4An AWS account is connected to a CSPM tool through a role that grants `ec2:Describe*`, `s3:GetBucket*` and `s3:ListBucket`, but not `iam:GetPolicyVersion` or `kms:GetKeyPolicy`. The account holds 1,620 EC2 instances and S3 buckets combined, 240 IAM roles and 90 customer-managed KMS keys. The dashboard reports 1,620 resources evaluated at 100% compliant, and the security lead signs off on the account. What does that sign-off rest on?
MCQ-25A 250-account AWS estate is scanned by a CSPM tenant through a read-only cross-account role in each account, while a separate auto-remediation role in each account can modify resources and is assumed by a Lambda in the security account. Coverage is reported as console-listed accounts with a successful scan in the last 24 hours, divided by all accounts listed in the console. Account 3312 ran its last workload on 3 March and completed migration; its provider contract keeps the account open and billable until 30 April, and it cannot be assumed into after that. The security standard requires any role that can change resources to be revoked within 7 days of an account's last workload terminating. The tenant retains findings for 12 months, matching the audit policy, but deleting an account record purges its findings immediately. It is 4 March. Which plan for account 3312 is correct?
MCQ-15A CSPM scan of 900 AWS accounts returns 12,480 findings. Grouping by root cause collapses them to 340 causes. The largest cause: a Terraform module pinned at v2.3.1 that creates an S3 bucket with a public-read ACL: accounts for 5,900 findings across 410 accounts. Hand-remediating one finding averages 20 minutes. Bumping the module and re-applying it across the 410 accounts averages 6 engineer-hours. To the nearest engineer-hour, how many engineer-hours does the module bump save compared with hand-remediating the 5,900 findings it accounts for?
MCQ-14A platform team runs a CSPM tool across 1,200 AWS accounts. In 90 days it raised 340 findings; 78% come from six rules, and 91% of those trace back to four Terraform modules in a shared registry that 60 service teams consume. The module pipeline runs `terraform plan` and `terraform apply` with no policy step, and developers merge module changes directly to main. Which change is most likely to stop the six rules from recurring?
MCQ-5A cloud security team has 2,400 open findings and capacity to remediate one this sprint. Its triage rubric scores a finding by likelihood of exploitation (internet reachability, authentication required, public exploit available) and by impact (what the compromised identity or data can reach). The team is comparing five findings: | ID | Finding | CVSS | Exposure | Public exploit | What a compromise reaches | | --- | --- | --- | --- | --- | --- | | F-1 | RCE in an internal CI runner | 9.8 | Internal only; no inbound route from the internet | No | No cloud credentials; no production data | | F-2 | SSRF in an API gateway | 6.5 | Internet-facing; endpoint requires no authentication | Yes | IAM role that can read the production customer database | | F-3 | Local privilege escalation on a bastion host | 8.1 | Internet-facing SSH, but a hardware key and the corporate VPN are required | No | No application service; no data | | F-4 | Public-read S3 bucket | 7.5 | Internet-facing | No | Published marketing PDFs | | F-5 | SQL injection in an internal admin panel | 8.6 | Internal only; VPN and SSO required | No | Panel can read the production customer database | Which finding should the team remediate first?
MCQ-6A graph-based posture tool enumerates seven attack paths that reach `rds-prod`, the database holding customer PII. Each path is a chain of hops: 1. `alb-public` → `web-01` → `role-web` → `rds-prod` 2. `alb-public` → `web-01` → `role-api` → `rds-prod` 3. `bastion` → `web-01` → `role-web` → `rds-prod` 4. `ci-runner` → `web-01` → `role-api` → `rds-prod` 5. `bastion` → `web-02` → `role-web` → `rds-prod` 6. `ci-runner` → `lambda-deploy` → `role-api` → `rds-prod` 7. `vpn-gateway` → `web-01` → `role-web` → `rds-prod` `rds-prod` must keep serving the application, so it cannot be changed or taken offline. Which single remediation breaks the most of these paths?
MCQ-20A CSPM tool reports two findings on the same production bucket, which holds customer PII and is reachable from the internet. One is static: default encryption has been disabled since the bucket was created 40 days ago. The other is drift: the public access block was turned off 20 minutes ago by a principal that is not one of the deployment pipeline's service accounts, and that setting is applied outside the bucket's IaC stack. Both findings carry the same severity. Which triage order is right?
MCQ-19A platform team's CSPM tool checks deployed resources against a vendor control set that is refreshed weekly. They want that same control set to block any Terraform merge that violates it, before apply. Their modules read data sources and rely on provider defaults, so several control-relevant attributes are not present in the HCL source. Which integration should they adopt?
MCQ-18A platform team runs a CSPM that ships a 380-control benchmark and already routes findings into the security team's ticketing queue with per-control owners and remediation text. An auditor requires that every S3 bucket whose name ends in `-prod` carry a `data-classification` tag whose value is one of `public`, `internal`, `confidential`, or `restricted`; any bucket missing the tag or carrying another value must appear as a high-severity finding in that same queue within 24 hours of creation. The team has no headcount to maintain new code and no budget for another scanner. Which approach meets the requirement?
MCQ-10A CSPM scan flags `s3-bucket-versioning-enabled` on `acme-temp-landing-prod` because versioning is disabled. The bucket receives files from a partner and a lifecycle rule deletes all objects after 7 days. The team requests an exception for this bucket and 14 other temporary buckets for 12 months. The product security lead can approve. What should the security engineer do?
MCQ-24A cloud security team reports quarterly to a CISO who wants one number showing whether the estate is getting safer. The total finding count rose from 3,200 to 5,100 after 15 new AWS accounts were onboarded, even though 400 critical findings were remediated. The CISO rejected the vendor secure score because it rose last quarter when a new policy set was enabled, not because exposure fell. The team has full CSPM connector coverage and a 14-day SLA for critical findings. Which metric should they report?
MCQ-12A CSPM deployment covers 300 accounts and reports 1,400 open findings. After an auto-remediation run deleted an IAM role that a quarterly job assumed, the team adopted a rule: enforce auto-remediation only for a finding class whose fix is reversible by a single API call and whose 30-day preview run logged no dependency breakages. The preview data: | finding class | open findings | fix | reversible by one API call | breakages logged | | --- | --- | --- | --- | --- | | S3 bucket without default encryption | 780 | PutBucketEncryption | yes | 0 | | Security group allows 0.0.0.0/0 on port 22 | 140 | revoke the rule | yes | 3 | | Unused IAM role with an admin policy attached | 70 | DeleteRole | no | 0 | | CloudTrail trail with logging stopped | 410 | StartLogging | yes | 0 | How many findings qualify for enforced auto-remediation?
MCQ-22A CSPM platform auto-remediates findings across 60 AWS accounts using one service principal in the AWS Organizations management account that holds AdministratorAccess. Last quarter a rule meant to close public S3 buckets instead applied a deny-all bucket policy to a customer-facing bucket, taking it offline for 40 minutes, and the CloudTrail record of the change was written by that same principal, which also holds cloudtrail:DeleteTrail. The team wants to keep unattended remediation for low-risk findings. Which change addresses both failures the incident exposed?
MCQ-13An automated remediation sets all four Block Public Access flags on an S3 bucket whose images are served by a CloudFront distribution. Until then the images were readable anonymously through the objects' public-read ACLs; the distribution has no origin access control or origin access identity and fetches objects over the bucket's REST endpoint. The job's read-back shows all four flags set. Ten minutes later, 403s on /images/* have gone from 0.1% to 31% of roughly 40,000 requests per minute, while image requests served from cached edge copies still return 200. What is the correct next step?
MCQ-23At 09:14 a CSPM scanner flags production bucket cust-exports: at 08:41 its bucket policy was rewritten to grant s3:GetObject on the raw/ prefix to any principal (Principal *). That prefix carries the tag data-classification: customer-pii. Account-level Block Public Access was disabled in Q2 under approved change CHG-4471. CloudTrail data events for the bucket show 12,140 GetObject requests from 203.0.113.47 in the 20 minutes after the rewrite: 12,100 returned 404 NoSuchKey and 40 returned 200, moving 6.2 GB out. Which action does this evidence warrant?