Explore / Google Cloud Professional Cloud Security Engineer Exam Prep / Practice questions
Google Cloud Professional Cloud Security Engineer practice questions
10 questions from the bank behind this course's timed mock. Free to read and to answer, no account needed, and each has its answer and an explanation. They belong to Google Cloud Professional Cloud Security Engineer Exam Prep.
Answer them one at a time →Read lesson 1 free
Question 1
A cloud engineer books the Professional Cloud Security Engineer exam early "to see what it's like" and fails. They retake it soon after and fail again. Under Google's program policies for Professional exams as covered in this course, what happens next?
- They can retake straight away, because waiting periods only apply to Associate exams
- They must wait 365 days, because a second failure triggers the longest waiting period
- They must wait 14 days, and the third attempt is free because the first two were failed
- They must wait 60 days before a third attempt and pay the full fee again; four attempts are allowed in two years
Show the answer
Answer: D. They must wait 60 days before a third attempt and pay the full fee again; four attempts are allowed in two years
The wait grows with each failure: 14 days after the first, 60 after the second and 365 after the third. Every attempt is paid for, and the limit is four attempts in two years. The 365-day option puts that longest wait one failure too early, after the second fail instead of the third.
Question 2
A security engineer at a manufacturing company must let a supplier's engineers use Google Cloud with the supplier's own Okta tenant, without creating any Google accounts for them. The engineer has the Owner role on the project 'partner-sandbox' and no roles at the organization level. Which TWO things must happen before the workforce pool and its provider can be configured? (Select TWO.)
More than one answer is right.
- Create the workforce pool inside the 'partner-sandbox' project with --location=us-central1
- Run Google Cloud Directory Sync to copy the supplier's users into Cloud Identity
- Give the engineer Workforce Identity Pool Admin (roles/iam.workforcePoolAdmin) on the organization
- Create the workforce pool at the organization level with --location=global
- Give the engineer Workload Identity Pool Admin on the 'partner-sandbox' project
Show the answer
Answers: C. Give the engineer Workforce Identity Pool Admin (roles/iam.workforcePoolAdmin) on the organization; D. Create the workforce pool at the organization level with --location=global
Workforce pools sit at the organization level with location global, so the engineer needs roles/iam.workforcePoolAdmin on the organization, and project Owner is not enough. A Workload Identity Pool Admin role is for workload federation, not human users, and running GCDS would copy identities into Google, which is the thing the requirement rules out.
Question 3
A security engineer sets the organization policy constraint constraints/storage.uniformBucketLevelAccess at the organization node to enforce IAM-only access. A week later, an audit finds that several buckets created last year still have object ACLs that grant access to external users. The goal is to remove ACL-based access from these buckets without touching their IAM policies. What should the engineer do?
- Set the same constraint again at each project level, because organization-level constraints do not apply to buckets
- Wait for the organization policy to finish propagating, because it will switch existing buckets to uniform bucket-level access automatically
- Remove allUsers and allAuthenticatedUsers from each bucket's IAM policy so that the ACLs can no longer grant access
- Enable uniform bucket-level access on each existing bucket, for example with gcloud storage buckets update --uniform-bucket-level-access
Show the answer
Answer: D. Enable uniform bucket-level access on each existing bucket, for example with gcloud storage buckets update --uniform-bucket-level-access
The constraint only applies to buckets created after it is set, so existing buckets keep their current setting until you turn on uniform bucket-level access for each one. Waiting for propagation won't help, because the policy never changes existing buckets. Editing IAM won't help either: in fine-grained mode an ACL grants access on its own.
Question 4
A security engineer has built a threat-prevention security profile and placed it in a security profile group. A firewall policy rule already sends web traffic to a zonal firewall endpoint using apply_security_profile_group. The CISO now wants exploits rated HIGH or CRITICAL to be blocked. Every lower-severity signature should keep its built-in default action. What should the engineer do?
- Create a second threat-prevention profile that denies HIGH and CRITICAL, and add it to the same security profile group
- Add a higher-priority firewall policy rule with the deny action on tcp:80 and tcp:443 for the web tier
- Add a severity override to the existing threat-prevention profile for HIGH and CRITICAL with the action DENY
- Add a severity override to the existing threat-prevention profile for HIGH and CRITICAL with the action ALERT
Show the answer
Answer: C. Add a severity override to the existing threat-prevention profile for HIGH and CRITICAL with the action DENY
Each signature has a default action, and a severity override with DENY changes only HIGH and CRITICAL, so the other severities keep their defaults. A second profile is not possible because a security profile group holds at most one profile of each type, and ALERT would only report the threats without blocking them.
Question 5
Your company is moving the public zone shop.example.com from Cloud DNS to another DNS provider. DNSSEC is currently on for the Cloud DNS zone, and its DS record is published at the registrar. The security engineer must turn DNSSEC off as part of the move without causing failed lookups for customers whose resolvers validate DNSSEC. What should they do first?
- Leave the DS record in place, because validating resolvers will treat the zone as unsigned once signing is off
- Run gcloud dns managed-zones update shop-public --dnssec-state off, then remove the DS record at the registrar
- Turn off signing and publish a new DS record at the registrar at the same time
- Remove the DS record at the domain registrar and let resolvers stop using it before turning off signing on the zone
Show the answer
Answer: D. Remove the DS record at the domain registrar and let resolvers stop using it before turning off signing on the zone
The DS record has to come out first. If signing stops while the DS record is still published, validating resolvers expect signatures that no longer exist, and every lookup for the domain fails. Turning off signing first is the exact order that causes the outage you are trying to avoid.
Question 6
You are a security engineer at a company with on-premises application servers. They connect to a production VPC over HA VPN. The servers must reach Cloud Storage through an internal IP address that your company chooses and owns, so the network team can advertise it over BGP and add it to on-premises firewall rules. Calls to Google APIs that VPC Service Controls does not support must be blocked. What should you configure?
- Reserve a global internal address outside every subnet range with purpose PRIVATE_SERVICE_CONNECT, then create a global forwarding rule that targets the vpc-sc Google APIs bundle.
- Create a private Cloud DNS zone for googleapis.com that points *.googleapis.com to restricted.googleapis.com, and add a route for 199.36.153.4/30 through the default internet gateway.
- Reserve an internal address inside the app subnet's primary range, then create a Private Service Connect forwarding rule that targets the vpc-sc Google APIs bundle.
- Reserve a global internal address outside every subnet range with purpose PRIVATE_SERVICE_CONNECT, then create a global forwarding rule that targets the all-apis Google APIs bundle.
Show the answer
Answer: A. Reserve a global internal address outside every subnet range with purpose PRIVATE_SERVICE_CONNECT, then create a global forwarding rule that targets the vpc-sc Google APIs bundle.
A Private Service Connect endpoint gives you an internal IP you choose, which you can advertise to on-premises, and the vpc-sc bundle blocks APIs that VPC Service Controls doesn't support, just like the restricted VIP. The restricted VIP also blocks those APIs, but it uses Google's fixed shared range 199.36.153.4/30, not an address your company owns. The all-apis bundle would let the unsupported APIs through.
Question 7
A platform engineer creates a key in the central project kms-prod and a bucket in project app-prod with that key as its default encryption key. The engineer's own account has roles/cloudkms.cryptoKeyEncrypterDecrypter on the key, but every upload to the bucket fails with a permission error. What should you do to fix the uploads with the least privilege?
- Grant roles/cloudkms.cryptoKeyEncrypterDecrypter on the key to the Cloud Storage service agent of app-prod
- Set constraints/gcp.restrictCmekCryptoKeyProjects to allow projects/kms-prod
- Recreate the key in app-prod so that the key and the bucket are in the same project
- Grant roles/cloudkms.admin on the key ring to the engineer's account
Show the answer
Answer: A. Grant roles/cloudkms.cryptoKeyEncrypterDecrypter on the key to the Cloud Storage service agent of app-prod
The Cloud Storage service agent of the data project uses the key to wrap and unwrap data, not the user, so the service agent needs Encrypter/Decrypter on the key. Moving the key into app-prod is the closest distractor, but it isn't needed: keys in a separate project work fine, and keeping them separate is how you get separation of duties.
Question 8
A healthcare ML team fine-tunes a model on patient records using NVIDIA H100 GPUs on Compute Engine. The data is already encrypted at rest with CMEK keys in Cloud HSM. Compliance now requires that the data also be protected while the GPU workload processes it, including memory-integrity protection against a compromised host. Which approach best meets this requirement?
- Create an a3-highgpu-1g VM with --confidential-compute-type=TDX and --maintenance-policy=TERMINATE
- Keep the existing A3 VM and add Cloud HSM-backed CMEK to every attached persistent disk and the boot disk
- Create an n2d-standard VM with --confidential-compute-type=SEV_SNP and --min-cpu-platform="AMD Milan", then attach the H100 GPUs
- Create a G4 VM with --confidential-compute-type=SEV so that the GPU host's memory is encrypted
Show the answer
Answer: A. Create an a3-highgpu-1g VM with --confidential-compute-type=TDX and --maintenance-policy=TERMINATE
Intel TDX on A3 High (H100) is the option for Confidential Computing on attached GPUs, and TDX adds integrity protection against a hostile host. TDX VMs can't live-migrate, so TERMINATE is the right maintenance policy. G4 with SEV is a real Confidential VM option, but SEV encrypts memory without SNP/TDX-style integrity protection, and CMEK only protects data at rest.
Question 9
A security engineer is auditing a reporting job's image in Artifact Registry. The image was built and pushed five months ago and hasn't been pulled since. Artifact Analysis shows no critical vulnerabilities for it, yet the same base image version has several critical CVEs published against it. The team wants a reliable, low-effort way to keep this image's vulnerability posture accurate. What is the best explanation and fix?
- Enrol the GKE cluster that runs the job in the Rapid release channel so that node image upgrades refresh the OS packages inside the container.
- Continuous analysis stops updating vulnerability data for images not pulled in the last 30 days, so the result is stale. Schedule a manual Cloud Build trigger with Cloud Scheduler to rebuild the image, which is then scanned on push and gated by a severity check.
- Artifact Analysis rescans every image in a repository daily whether or not it is used, so the result is current and the published CVEs probably don't affect this image.
- Create a Pub/Sub trigger on the gcr topic, filtered to this image's own tag, so Cloud Build rebuilds the image whenever its vulnerability data changes.
Show the answer
Answer: B. Continuous analysis stops updating vulnerability data for images not pulled in the last 30 days, so the result is stale. Schedule a manual Cloud Build trigger with Cloud Scheduler to rebuild the image, which is then scanned on push and gated by a severity check.
Continuous analysis only refreshes data for images pulled in the last 30 days, so an unused image can look clean. A scheduled rebuild produces a fresh image that is scanned on push. The gcr topic option fails because Artifact Registry publishes there when images are pushed, tagged or deleted, not when vulnerability data changes.
Question 10
A retail company is moving a workload from GKE Standard to GKE Autopilot. The platform lead tells the security engineer: "On Autopilot, Google manages the nodes, so our cluster security work is done." Which TWO responsibilities stay with the customer after the move? (Select TWO.)
More than one answer is right.
- Upgrading the kubelet and other managed node components to the latest supported version
- Maintaining Kubernetes RBAC and IAM policies that control who can deploy to and read from the cluster
- Hardening and patching the Container-Optimized OS image running on the worker nodes
- Rotating cluster credentials and making sure logging and monitoring are enabled, then responding to the alerts
- Operating and backing up the cluster's etcd datastore and API server certificates
Show the answer
Answers: B. Maintaining Kubernetes RBAC and IAM policies that control who can deploy to and read from the cluster; D. Rotating cluster credentials and making sure logging and monitoring are enabled, then responding to the alerts
Even on Autopilot, the customer still owns access policy (RBAC/IAM), credential rotation, enabling logging and monitoring, and responding to alerts. Node OS hardening, the control plane (etcd, certificates) and managed node components such as the kubelet are Google's job, so those options are real tasks that sit on Google's side.
The full timed mock
50 questions in 120 minutes, with a pass mark of 70%. Each paper is drawn from a bank of questions written for the mock alone, and these 10 are left out of it. Full access opens it, with every level of the course.
LevelUpWith is independent of the organisation that sets this exam. These questions were written for this course; they are not taken from the exam.