GCP-PSE Study Guide: Every Security Control Answers One Question, So Learn the Questions
The Professional Cloud Security Engineer exam has a lot of products in it, and many of them sound alike: policies, perimeters, boundaries, proxies, posture. The way through is to notice that every control answers exactly one question. IAM answers "who may call this API?". VPC Service Controls answers "may this data cross this line?". Access Transparency answers "what did Google staff do?". Wrong options in this exam are almost always a real control answering a different question. This guide groups the exam by those questions.
Study guide · Google Cloud Certified Professional Cloud Security Engineer (GCP-PSE)

The exam
50 to 60 multiple choice questions in 2 hours. Google does not publish a pass mark; check the official exam page for current details. Five sections, with the exam guide's approximate weights: Configuring access (~25%), Securing communications and establishing boundary protection (~22%), Ensuring data protection (~23%), Managing operations (~19%) and Supporting compliance requirements (~11%).
Of the 275 GCP-PSE questions we wrote for this question bank, 79 (29%) are troubleshooting questions: something is blocked that should work, or allowed that should not. Service accounts appear in the stem or options of 45 questions, organisation policies in 35 and Cloud KMS in 23. If you can explain why a request was allowed or denied, layer by layer, you can answer a large part of this exam.
The question-to-control map
| The question | The control that answers it |
|---|---|
| Who is this person, and how did they sign in? | Cloud Identity with SAML SSO; Workforce Identity Federation |
| What is this software, running outside Google Cloud? | Workload Identity Federation (no keys) |
| Which API calls may this principal make? | IAM allow policies, conditions; deny policies on top |
| Which configurations are allowed at all? | Organization Policy (built-in and custom constraints) |
| Can data leave these projects, even with valid credentials? | VPC Service Controls perimeter |
| Which packets may reach this VM? | Cloud NGFW firewall policies and rules |
| Who may open this web app, from which device? | Identity-Aware Proxy with access levels |
| What does an attacker on the internet reach first? | Cloud Armor at the load balancer |
| What is readable if this data leaks? | Sensitive Data Protection (de-identify); CMEK (revocable keys) |
| What may run in this cluster? | Binary Authorization |
| What did our people do? What did Google's people do? | Cloud Audit Logs; Access Transparency (Access Approval gates Google's access) |
"Who are you?": identity
A company that keeps Active Directory needs two separate flows. Provisioning creates users and groups in Cloud Identity, usually with Google Cloud Directory Sync. Sign-in sends the user to the identity provider through SAML, where the password and MFA are checked. Questions mix the two on purpose: a sync problem is not fixed by changing SSO, and an SSO failure where the assertion's NameID does not match the user's primary email is not fixed by re-syncing. Super admin accounts bypass the external identity provider, so protect them with Google's own 2-step verification.
For software, the rule is "no keys if there is any alternative": attached service accounts inside Google Cloud, Workload Identity Federation outside it, impersonation for humans who need to act as an account. Block new keys with an organisation policy, and do not leave the Editor role in projects that still allow key creation, because Editor can create a key for a more powerful account in the same project.
"What may you do?": authorisation
Allow policies grant and are inherited. Deny policies are checked first, are inherited, and are the way to set one central guardrail (for example: nobody except a named group may delete projects tagged as production). Privileged Access Manager replaces standing admin rights with time-limited grants that can require a justification and an approval. And when something does not behave as expected, use the right diagnostic: Policy Analyzer answers "who has access to what", Policy Troubleshooter answers "why does this principal have, or lack, this permission", and Policy Simulator answers "what would change if I applied this policy".
"Can it cross this line?": boundaries
Match the control to the layer. Firewall policies work on IP addresses and ports. Cloud Armor filters HTTP(S) traffic at the edge, before the backend. IAP gates a web application by identity and context. VPC Service Controls works at the level of Google APIs: it stops data moving between projects inside the perimeter and anything outside it.
Going deeper: why a firewall cannot stop data leaving through a Google API
Every Cloud Storage bucket in the world, including one owned by an attacker, is reached through the same endpoint, storage.googleapis.com. A firewall rule that allows a VM to talk to Cloud Storage therefore allows it to copy data to any bucket, and a rule that blocks the endpoint breaks legitimate work too. IAM does not help either when the caller holds valid credentials, which is exactly the case with stolen keys or an insider. VPC Service Controls looks at the request itself: which project the target resource belongs to, and whether that project is inside the perimeter. A copy to a bucket outside is denied even though the network path and the IAM grant are both fine. That is why Google recommends using IAM and a perimeter together, and why changes to a perimeter should be tested in dry run mode first: dry run logs what would be blocked without blocking it.
Beginner tip: in two test projects, create a dry-run VPC Service Controls perimeter around one of them, give a VM's service account in the protected project write access to a bucket in the other, and copy a file there. The copy succeeds, but the audit logs now show a dry-run violation that names the perimeter and the reason. Reading one real violation entry makes the ingress and egress rule questions far easier, because the entry shows exactly which identity, method and project the rule would have to match.
"What is readable if it leaks?": data protection
Google Cloud already encrypts data at rest by default. CMEK does not add missing encryption; it adds control: where the key lives, how it rotates, an audit trail of key use, and the ability to disable the key and cut the service off from the data. It does not stop an authorised user who reads through the service, because the service decrypts on the caller's behalf. Cloud HSM keeps the key in hardware; Cloud EKM keeps it outside Google, at the price of depending on the external manager's availability.
Sensitive Data Protection discovers, inspects and de-identifies. Know which transformations can be reversed: redaction and masking cannot; deterministic encryption (AES-SIV) and format-preserving encryption can, with the same cryptographic key (typically wrapped by Cloud KMS).
"How will we know?": operations and evidence
Admin Activity audit logs are always on; Data Access logs are off except for BigQuery, so a detector that "sees nothing" is often missing them. Firewall rules logging records the decision when a connection starts; VPC Flow Logs sample ongoing traffic. Neither records payload; Packet Mirroring does. Route logs with an aggregated sink to a locked central project. For supply chain questions: vulnerability scanning finds issues, Binary Authorization enforces attestations at deploy time.
"What did the provider do?": compliance
Access Transparency records what Google personnel did with your data; Access Approval requires your approval before they get access (for supported services, with documented exclusions). Assured Workloads applies a control package to a folder for a regulatory regime. No single product satisfies a regulation by itself, and new organisation policies are usually not retroactive: pair a preventive constraint with a detective control that finds existing violations.
A study order
- Identity and IAM (~25%), with the diagnostic tools, since troubleshooting questions start here.
- Boundaries: firewall policies, Cloud Armor, IAP, then VPC Service Controls in depth.
- Data protection: Cloud KMS, CMEK, Sensitive Data Protection, Secret Manager.
- Logging, detection and supply chain, then compliance controls, then mixed sets.
Practise
Practise in the question bank: every question explains each option, right and wrong. Start with the domain you just studied, then take a timed mock exam.
- GCP-PSE exam page · All 335 GCP-PSE questions · Timed mock exam · 23-day study plan
- By domain: Configuring access · Securing communications and establishing boundary protection · Ensuring data protection · Managing operations · Supporting compliance requirements
Approximate section weights as listed in Google's Professional Cloud Security Engineer exam guide. Security product names and tiers change often; check the current official exam page and documentation.