TL;DR
Certification earned
After DVA-C02, SOA-C03, DOP-C02, and SAP-C02, SCS-C03 brought the same AWS services into a security-focused conversation.
The question is not simply “which service detects this?” It is usually closer to:
Which control prevents, detects, contains, or proves this requirement at the correct boundary?
I organized these notes around the official SCS-C03 domains. Each domain contains the rules I reviewed, the distinctions that caused the most mistakes, and original practice questions. These are study exercises, not reproduced exam questions.
Exam Overview and Domain Breakdown
The AWS Certified Security – Specialty exam covers six primary domains:
| Domain | Description | Exam Weighting |
|---|---|---|
| Domain 1 | Detection | 16% |
| Domain 2 | Incident Response | 14% |
| Domain 3 | Infrastructure Security | 18% |
| Domain 4 | Identity and Access Management | 20% |
| Domain 5 | Data Protection | 18% |
| Domain 6 | Security Foundations and Governance | 14% |
The exam consists of 65 questions: 50 scored questions and 15 unscored pretest questions. The time limit is 170 minutes, and the passing score is 750 out of 1,000. It can include multiple-choice, multiple-response, ordering, and matching questions.
The domain order I used while studying was:
Check the official AWS certification page before scheduling. Exam blueprints and service capabilities change, so the current exam guide should take priority over old notes.
My Study Workflow: One Guide, Two Cheat Sheets
I used three study layers:
| Study Layer | Question It Answers |
|---|---|
| Full domain notes | What does this service or control do? |
| High-impact cheat sheet | Which distinction or limitation is easy to forget? |
| Question-and-answer review | Why is this option correct, and why do the alternatives fail? |
For every service, I wrote four things:
- Control boundary: principal, resource, network, data, account, or organization.
- Control timing: before an API call, during traffic flow, after a configuration change, or during an incident.
- Important limitation: what the service cannot inspect, prevent, or administer.
- Native integration: which managed service routes, aggregates, or automates the result.
My final review loop was:
The most useful notes explained why a tempting answer was incomplete. For example, an S3 bucket policy can allow a cross-account read while the request still fails because the KMS key policy does not allow decryption.
Domain 1: Detection (16% Weighting)
Domain 1 focuses on collecting security telemetry, identifying threats and vulnerabilities, and routing findings to the right investigation or response workflow.
1. Match the Finding to the Detection Service
| Service | Primary Purpose | Important Boundary |
|---|---|---|
| Amazon GuardDuty | Threat detection from AWS telemetry, including suspicious activity and malware-related findings | Findings are routed through EventBridge; GuardDuty is not a general vulnerability scanner. |
| Amazon Inspector | Continuous vulnerability scanning for EC2, ECR, and Lambda | Finds software and package vulnerabilities, not every active compromise. |
| Amazon Macie | Discover sensitive data in S3 | It is focused on S3 data discovery and classification. |
| Amazon Detective | Investigate findings with entity relationships and timelines | It supports investigation; it is not the primary containment mechanism. |
| AWS Security Hub | Aggregate and normalize findings into ASFF | It centralizes findings but does not replace the source service’s telemetry. |
| Amazon Security Lake | Centralize security logs using OCSF | It is a security data lake, not an inline prevention control. |
The exam often presents a real service in the wrong role. Inspector does not replace GuardDuty for threat detection. Security Hub does not reconstruct every API call by itself. Macie discovers sensitive data; it does not become an application authorization layer.
2. CloudTrail, Config, and Network Telemetry
- CloudTrail management events record control-plane API activity such as changing a bucket policy or disabling a trail.
- CloudTrail data events record object-level or function-level activity such as
PutObjector LambdaInvoke; they must be configured for the required resources and Regions. - AWS Config records resource configuration and compliance state. It is useful for answering what a resource looked like and when it changed.
- VPC Flow Logs provide network metadata, not packet payloads. Use a custom format when the scenario requires fields missing from the default format.
- Route 53 Resolver query logs help investigate suspicious DNS behavior and possible command-and-control activity.
- Transit Gateway Flow Logs provide visibility into traffic crossing a centralized network hub.
CloudTrail answers who called which AWS API. Config answers what the resource configuration was. Flow Logs answer which network connection metadata was observed. I kept these questions separate during review.
3. Centralize Findings Without Centralizing Everything
For a multi-account organization, I used a dedicated Security account as the delegated administrator for supported security services and a separate Log Archive account for protected logs.
| Requirement | Pattern |
|---|---|
| Aggregate GuardDuty and Inspector findings | Security Hub administrator and member accounts |
| Route critical findings to automation | EventBridge pattern to SSM Automation or Lambda |
| Send notifications | EventBridge to SNS, User Notifications, or an approved collaboration target |
| Retain normalized security logs | Security Lake with S3 lifecycle and subscriber access |
| Protect audit records | Log Archive account with restricted access, Object Lock, and suitable KMS protection |
The management account remains an Organizations and billing boundary, not the default home for workloads and security evidence.
Domain 1 Practice Questions
Question 1
A security team wants to receive an email when GuardDuty detects an unauthorized access-key finding. Which design uses the native event path?
- A. Configure a CloudWatch alarm on the GuardDuty console.
- B. Create an EventBridge rule for the finding and target an SNS topic.
- C. Configure GuardDuty to publish directly to SNS.
- D. Poll the GuardDuty API from a scheduled EC2 instance.
Show answer and reason
Answer: B.
Reason: GuardDuty findings can be matched by EventBridge and routed to SNS or automation. GuardDuty does not publish directly to SNS, and polling adds unnecessary operational overhead.
Question 2
An investigation needs to identify who called PutBucketPolicy and when the bucket policy changed. Which service combination provides the required evidence?
- A. CloudTrail management events and AWS Config configuration history.
- B. VPC Flow Logs and Amazon Inspector.
- C. Macie sensitive-data findings and Security Hub.
- D. Route 53 Resolver query logs and GuardDuty.
Show answer and reason
Answer: A.
Reason: CloudTrail identifies the API caller and event time. Config can show the resource configuration history and help compare the policy state before and after the change.
Question 3
A company wants to find a custom account-number pattern in millions of existing S3 objects and apply tags to the matching objects without downloading them. Which design fits?
- A. Macie custom data identifier, EventBridge, Lambda, and S3 Batch Operations.
- B. GuardDuty Runtime Monitoring and an EC2 patch baseline.
- C. Inspector enhanced scanning and an ECR lifecycle policy.
- D. Security group rules and VPC Traffic Mirroring.
Show answer and reason
Answer: A.
Reason: Macie discovers sensitive content, while S3 Batch Operations can process many objects in place. EventBridge and Lambda can coordinate tagging or follow-up actions.
Domain 2: Incident Response (14% Weighting)
Domain 2 focuses on containing incidents, preserving evidence, investigating root cause, and automating repeatable response actions.
1. Contain First, Then Preserve Evidence
My incident-response sequence was:
Detect → contain → preserve → investigate → eradicate → improve
For a compromised EC2 instance:
- Restrict network access without destroying the instance.
- Prevent the instance from being returned to service by an Auto Scaling or load-balancer workflow.
- Capture EBS snapshots and required host or memory evidence.
- Investigate CloudTrail activity, network telemetry, and related identities.
- Eradicate the compromise and rebuild from a trusted source.
- Improve the controls that allowed the incident.
Security groups are stateful and allow rules only. NACLs are stateless and support explicit deny rules, so a restrictive subnet NACL may be required when active connections must be stopped quickly.
2. Preserve IAM and Compute Evidence
For a compromised IAM user, deactivate access keys and attach a deny policy while preserving the user and audit trail. Do not delete the user immediately; deletion can remove context and makes rollback harder if the incident is later found to be accidental.
For compute forensics:
- Capture EBS snapshots before termination or restoration.
- Preserve relevant CloudTrail events and security findings.
- Capture volatile evidence when the scenario explicitly requires memory or live-host state.
- Use SSM Run Command or an approved forensics workflow instead of opening emergency SSH access.
- Record every containment action and its timestamp.
3. Event-Driven Response Automation
The common automation pattern is:
GuardDuty / Config / CloudTrail finding
↓
EventBridge rule
↓
SSM Automation or Lambda
↓
Isolate → snapshot → notify → record
Prefer a managed SSM Automation runbook when an AWS-native action already exists. Use Lambda when the response requires custom orchestration, such as identifying the affected Auto Scaling group, deregistering a target, taking snapshots, and writing an incident record.
Response automation should be narrow and idempotent. A rule that terminates every instance with a broad environment tag may create a second incident.
Domain 2 Practice Questions
Question 1
GuardDuty identifies a compromised EC2 instance. The security team must stop outbound traffic and preserve evidence. What should happen first?
- A. Terminate the instance and restore the latest AMI.
- B. Apply restrictive network controls and capture forensic snapshots.
- C. Reboot the instance and run Patch Manager.
- D. Delete the instance profile role.
Show answer and reason
Answer: B.
Reason: Contain the threat while preserving evidence. Termination, reboot, and restoration can destroy the state needed for investigation.
Question 2
An access key belonging to an IAM user launched unauthorized EC2 instances. The team is not yet sure whether a production process also uses the key. What is the safest immediate action?
- A. Delete the IAM user and all associated resources.
- B. Deactivate the access key, attach a deny policy, and preserve CloudTrail evidence.
- C. Rotate the key and leave the old key active until testing is complete.
- D. Disable the AWS account root user.
Show answer and reason
Answer: B.
Reason: Deactivation contains the credential while keeping the identity and audit trail available. Deleting the user is unnecessarily destructive, and rotating without deactivating leaves the compromised credential usable.
Question 3
A GuardDuty finding should automatically isolate only the affected EC2 instance, preserve its EBS volumes, and notify responders. Which design has the least custom infrastructure?
- A. EventBridge rule invoking an SSM Automation runbook with an approval or notification step.
- B. A cron job that scans every EC2 instance every minute.
- C. A CloudWatch dashboard that responders check manually.
- D. A public Lambda endpoint called by the compromised instance.
Show answer and reason
Answer: A.
Reason: EventBridge provides event routing and SSM Automation provides managed, auditable operational steps. The runbook can be scoped to the finding’s instance and designed to preserve evidence before remediation.
Domain 3: Infrastructure Security (18% Weighting)
Domain 3 focuses on securing network paths, edge services, private connectivity, host access, and traffic inspection.
1. Choose the Control by Network Layer
| Requirement | Appropriate Control | Exam Boundary |
|---|---|---|
| HTTP/S request filtering | AWS WAF | Supports CloudFront, ALB, API Gateway, AppSync, and other supported resources; not NLB. |
| Layer 3–7 inline inspection | AWS Network Firewall | Requires correct routing and symmetric inspection behavior. |
| Passive packet inspection | VPC Traffic Mirroring and GWLB | Copies traffic to an inspection fleet; it is not the same as inline blocking. |
| Subnet-level explicit deny | Network ACL | Stateless; allow return traffic and respect rule-number precedence. |
| Instance-level stateful filtering | Security Group | Stateful and allow-only; no explicit deny rule. |
| Outbound domain restriction | Route 53 Resolver DNS Firewall | Controls DNS queries, not every connection to a hard-coded IP. |
The phrase block these IP addresses can point to different controls depending on the resource and direction. WAF is appropriate for HTTP/S resources. NACLs support explicit IP denies at a subnet boundary. Security groups cannot express a deny rule.
2. Private Connectivity and Traffic Encryption
- PrivateLink: Consumer-initiated access to a specific service through an interface endpoint. It avoids broad VPC routing and can work with overlapping CIDRs.
- VPC peering: Private routing between VPCs, but it does not support overlapping CIDR ranges.
- Transit Gateway: Centralized routing between attachments; route-table association and propagation determine isolation.
- Direct Connect: Dedicated connectivity, but not encryption by itself.
- Site-to-Site VPN over private connectivity: Adds IPsec encryption when the requirement is encrypted traffic over a private path.
- Gateway endpoints: S3 and DynamoDB route-table integration, generally used to avoid NAT for same-Region service access.
- Interface endpoints: Private ENIs with security groups, endpoint policies, and private DNS support.
3. Host Hardening and Secure Administration
- Use SSM Session Manager to administer supported instances without public SSH/RDP ingress or bastion hosts.
- Use SSM Run Command for fleet-wide scripts and configuration changes.
- Use EC2 Image Builder to create patched golden AMIs and enforce approved images in launch templates.
- Use Amazon Inspector for continuous vulnerability scanning of supported compute and package surfaces.
- Use VPC endpoints so private instances can reach required AWS services without unnecessary internet paths.
- Use restrictive security-group egress rules when a compromised host must not reach arbitrary destinations.
4. Network Inspection Decisions
| Requirement | Pattern |
|---|---|
| IP/port metadata and accepted/rejected flows | VPC Flow Logs |
| Deep inspection without placing the appliance inline | Traffic Mirroring to GWLB |
| Inline filtering and TLS inspection | AWS Network Firewall with correct routing |
| Block HTTP floods and malicious web requests | WAF managed rules, rate-based rules, and IP sets |
| Allow private access to one partner service | PrivateLink interface endpoint |
Domain 3 Practice Questions
Question 1
A company must block three IPv4 addresses from reaching an ECS service behind an ALB. Which control provides explicit deny rules at the service’s subnet boundary?
- A. Security group deny rules.
- B. Inbound NACL deny rules with lower rule numbers than broad allows.
- C. An S3 bucket policy.
- D. A Route 53 latency policy.
Show answer and reason
Answer: B.
Reason: NACLs support explicit denies and apply at the subnet boundary. Rule-number precedence matters because evaluation stops at the first match. Security groups support allow rules only.
Question 2
An EKS workload must consume a partner’s service privately. The VPC CIDR ranges overlap, and the partner must not initiate connections into the EKS VPC. Which solution fits?
- A. VPC peering with routes in both directions.
- B. Transit Gateway with full route propagation.
- C. An interface VPC endpoint connected to the partner’s PrivateLink endpoint service.
- D. A public load balancer with an IP allow list.
Show answer and reason
Answer: C.
Reason: PrivateLink provides consumer-initiated, service-specific connectivity without requiring general bidirectional routing, and it avoids the overlapping-CIDR problem.
Question 3
Security engineers need passive deep packet inspection of traffic from a fleet of EC2 instances without adding inline latency to the application path. Which architecture is appropriate?
- A. VPC Traffic Mirroring to a GWLB endpoint and inspection fleet.
- B. A gateway route through an internet-facing load balancer.
- C. CloudWatch Metrics with a CPU alarm.
- D. A security group deny rule on the application instances.
Show answer and reason
Answer: A.
Reason: Traffic Mirroring copies ENI traffic to an out-of-band inspection path. Flow Logs provide metadata only, while an inline route would change the application path.
Domain 4: Identity and Access Management (20% Weighting)
Domain 4 is the largest domain. It focuses on policy evaluation, federation, delegation, privilege boundaries, and access analysis.
1. Policy Evaluation and Guardrails
The evaluation checklist I used was:
- An explicit deny wins.
- SCPs limit principals in member accounts; they do not grant permissions.
- Resource control policies limit supported resources; they do not replace resource policies.
- Cross-account access normally requires an allow in the caller’s identity policy and the target resource policy.
- Permission boundaries cap the maximum permissions an identity-based policy can grant.
- Session policies narrow one STS session.
- Trust policies allow role assumption; they do not grant the role’s service permissions.
- KMS key policies authorize use of the KMS key and must align with the caller’s permissions.
The effective permission is not “whatever one policy allows.” It is the result of the applicable identity, resource, organization, boundary, session, and key-policy decisions.
2. IAM Delegation and Privilege Escalation
To let developers create roles without allowing unrestricted escalation:
- Require an approved permissions boundary during
iam:CreateRole. - Allow only the required policy-management actions.
- Deny removal of the approved boundary.
- Restrict which managed or customer-managed policies can be attached.
- Use CloudTrail and IAM Access Analyzer to review actual access.
For ABAC, match principal and resource tags and require request tags at creation time. Tag policies standardize supported tag values, but they do not automatically require every resource to have every tag or grant access.
3. Workforce, Customer, and Third-Party Identity
| Identity Requirement | AWS Pattern |
|---|---|
| Workforce SSO into many AWS accounts | IAM Identity Center with permission sets |
| Federation from an enterprise IdP | SAML 2.0 and SCIM with the appropriate identity source |
| Customer application registration and sign-in | Cognito User Pools |
| Temporary AWS credentials for an authenticated customer | Cognito Identity Pools |
| Third-party cross-account role assumption | STS AssumeRole with an external ID |
| Service-to-service access | IAM roles and temporary credentials |
I kept workforce identity and customer identity separate. IAM Identity Center is for workforce access to AWS accounts and applications. Cognito User Pools authenticate application users. Cognito Identity Pools can exchange trusted identities for temporary AWS credentials.
4. Find and Reduce Unused Access
- IAM Access Analyzer: External access findings, unused access analysis, policy validation, and policy generation from activity.
- IAM Access Advisor: Service last-accessed information for reviewing permissions.
- Credential reports: IAM user credential status, MFA, key age, and last-use information.
- CloudTrail activity: Evidence for what a principal actually called, subject to configured event coverage.
Do not confuse “this resource is shared externally” with “this role has unused permissions.” They are different analysis questions.
Domain 4 Practice Questions
Question 1
Developers may create IAM roles in a development account but must not create roles with permissions outside an approved boundary. Which control directly enforces this?
- A. Require the approved permissions boundary in the role-creation condition and deny its removal.
- B. Create a CloudWatch dashboard showing all role policies.
- C. Attach a security group to the developer’s workstation.
- D. Add an S3 bucket policy to the development account.
Show answer and reason
Answer: A.
Reason: A permissions boundary caps the maximum permissions of the created role. Monitoring can detect the issue later, but it does not prevent the privilege-escalation path.
Question 2
An external SaaS provider assumes a role in the customer’s account. The customer needs protection from a confused-deputy scenario when multiple customers use the same provider. Which trust-policy condition is appropriate?
- A. Require a provider-supplied
sts:ExternalId. - B. Require the customer’s source VPC ID in an S3 bucket policy.
- C. Add an SCP that grants the provider access.
- D. Make the role trust the provider’s root account without conditions.
Show answer and reason
Answer: A.
Reason: An external ID binds the provider’s role assumption to the intended customer context and helps mitigate confused-deputy risk. An SCP cannot grant the missing trust or role permissions.
Question 3
A company wants employees to sign in once and access multiple AWS accounts using centrally managed permissions. Which design is appropriate?
- A. IAM Identity Center with groups and permission sets.
- B. An IAM user with an access key in each account.
- C. Cognito Identity Pools without an enterprise workforce IdP.
- D. A shared root-user credential stored in Secrets Manager.
Show answer and reason
Answer: A.
Reason: IAM Identity Center provides workforce SSO and provisions account roles from permission sets. Long-lived IAM keys and root credentials are not appropriate for workforce access.
Domain 5: Data Protection (18% Weighting)
Domain 5 focuses on encryption, key management, secrets, data discovery, immutability, and protection in transit.
1. KMS Key Types and Rotation
| Key Scenario | Key Choice or Rule |
|---|---|
| Basic service encryption with no custom key-policy requirement | AWS-owned or AWS-managed key, depending on the service |
| Cross-account sharing or custom key policy | Customer-managed KMS key |
| Temporary delegated access at scale | KMS grants |
| Key use only through a service | kms:ViaService condition |
| Imported key material | Rotation and replacement behavior differs from KMS-generated material |
| New key ID on a schedule | Create and transition to new keys; automatic rotation alone keeps the key ID |
| Regional decryption requirements | Consider KMS multi-Region key design where supported |
Automatic rotation changes backing key material while retaining the KMS key ID. If the requirement explicitly says “a new key ID every month,” automatic rotation is not enough.
2. Cross-Account Encryption Authorization
For an encrypted S3 object read across accounts, I traced:
- The caller’s IAM policy allows the required S3 and KMS actions.
- The S3 bucket policy allows the intended principal or organization.
- The KMS key policy allows the caller account or role to use the key.
- The VPC endpoint policy and
aws:SourceVpcecondition match the actual path, if endpoint restriction is required. - The KMS encryption context and service path do not contradict the request.
An S3 allow is not a KMS allow. A KMS key policy is not a substitute for the caller’s missing cross-account identity permission.
3. S3 Discovery, Encryption, and Immutability
- Use Macie to discover PII and custom sensitive-data patterns in S3.
- Use S3 Batch Operations to process large object sets without application-side downloads.
- Use SSE-KMS when customer-managed key control and auditable key usage are required.
- Use S3 Bucket Keys when reducing KMS request volume is important.
- Enable versioning and Object Lock for WORM retention.
- Use Compliance mode when no principal, including root, may shorten retention or delete protected versions.
- Treat existing objects and future objects separately; changing a default does not automatically repair every historical object.
4. Secrets, Certificates, and Encryption in Transit
| Requirement | Appropriate Pattern |
|---|---|
| Database credentials requiring scheduled rotation | Secrets Manager rotation, including alternating users where needed |
| Configuration with lower-cost secure storage and manual/custom rotation | SSM Parameter Store SecureString |
| Internal TLS and mTLS certificates | ACM Private CA |
| Public certificates for integrated AWS services | ACM public certificates |
| Client-side attribute encryption | AWS Database Encryption SDK with suitable KMS key material |
| Key custody or specialized HSM requirements | CloudHSM or an appropriate KMS custom key store |
Do not put secrets in source code, container images, or plaintext environment variables. Use an application role to retrieve the secret at runtime and restrict the secret, KMS key, and rotation function.
Domain 5 Practice Questions
Question 1
An analytics role in Account B reads SSE-KMS encrypted S3 objects in Account A. The bucket policy already allows the role. Which additional configuration is required?
- A. Allow
kms:Decryptfor the role in its IAM policy and allow the role in the KMS key policy. - B. Attach an SCP granting
kms:Decrypt. - C. Enable S3 Transfer Acceleration.
- D. Add a CORS rule to the bucket.
Show answer and reason
Answer: A.
Reason: Cross-account KMS use requires the caller’s IAM authorization and the target key policy authorization. SCPs limit permissions; they do not grant them.
Question 2
An organization must prevent anyone, including the root user, from deleting selected S3 reports before a seven-year retention period expires. Which configuration is appropriate?
- A. Versioning and Object Lock Compliance mode with retention applied to the protected versions.
- B. Governance mode with administrators allowed to bypass retention.
- C. An SCP denying
s3:DeleteObject. - D. A lifecycle rule that deletes objects after seven years.
Show answer and reason
Answer: A.
Reason: Object Lock Compliance mode provides the required WORM behavior. An SCP or bucket policy does not provide equivalent root-user protection, and Governance mode is bypassable by authorized principals.
Question 3
An RDS application must rotate credentials every 45 days without downtime. The application cannot be changed to coordinate a shared password update. Which strategy best fits?
- A. Store the password in a Lambda environment variable and update it manually.
- B. Use Secrets Manager alternating-users rotation with the application retrieving the current secret.
- C. Store the password in an encrypted Git repository.
- D. Use an S3 object with a seven-year retention policy.
Show answer and reason
Answer: B.
Reason: Secrets Manager alternating-users rotation maintains two database users so one can be rotated while the other serves traffic. The application retrieves the current secret rather than embedding credentials.
Domain 6: Security Foundations and Governance (14% Weighting)
Domain 6 focuses on the shared responsibility model, multi-account security architecture, compliance evidence, and organization-wide controls.
1. A Secure Multi-Account Baseline
I used this account model while studying:
Management Account → Organizations, billing, Control Tower only
Log Archive Account → CloudTrail, Config, VPC Flow, WAF logs
Security Account → Delegated GuardDuty, Security Hub, Macie, Inspector
Network Account → Transit Gateway, DNS, Direct Connect
Workloads OU → Production, staging, and development accounts
The point is separation of duties:
- The management account should not become the workload or log archive account.
- The Log Archive account should restrict deletion and administrative access.
- The Security account should administer supported security services.
- Workload accounts should operate applications under OU guardrails.
2. Preventive, Detective, and Corrective Governance
| Control Type | Examples | What It Does |
|---|---|---|
| Preventive | SCP, RCP, declarative policy, permission boundary, key policy | Blocks or limits an action before or during authorization |
| Detective | Config rule, Config aggregator, Security Hub standard, CloudTrail analysis | Finds a state or action that violates a requirement |
| Corrective | SSM Automation remediation, EventBridge workflow, Lambda response | Changes a resource or contains an incident after detection |
| Evidence | Audit Manager, CloudTrail, Config history, AWS Artifact | Supports investigation, audit, or compliance evidence |
When a question says “must not be possible,” I selected a preventive control. When it says “identify noncompliant resources,” I selected a detective control. When it says “automatically fix,” I looked for a corrective workflow.
3. Central Security Services and Organization Deployment
- Use Security Hub Central Configuration or supported delegated-administrator patterns to standardize security settings.
- Use AWS Config aggregators for read-only cross-account visibility.
- Use Config conformance packs to deploy related compliance rules.
- Use Firewall Manager for centralized WAF, Shield, security-group, Network Firewall, and DNS Firewall policies where supported.
- Use service-managed CloudFormation StackSets with automatic deployment for baselines in current and future OU accounts.
- Use AWS RAM for supported resource sharing, such as VPC subnets; do not assume every AWS resource can be shared through RAM.
4. Compliance and Shared Responsibility
| Service | Purpose |
|---|---|
| AWS Artifact | AWS compliance reports, certifications, and agreements |
| AWS Audit Manager | Collect evidence for an organization’s audit framework |
| AWS Config | Configuration history and compliance evaluation |
| AWS Well-Architected Tool | Review workload design against architectural pillars |
| AWS Trusted Advisor | Checks across cost, performance, security, and service limits depending on support |
AWS is responsible for security of the cloud. The customer remains responsible for security in the cloud, with the exact split varying by service. Using a managed service reduces operational responsibility; it does not eliminate IAM, data classification, configuration, or access-control decisions.
Domain 6 Practice Questions
Question 1
An organization wants every new account in the Workloads OU to receive a baseline IAM role automatically. Developers must not be able to delete the baseline resources. Which combination is appropriate? (Select TWO)
- A. Use a service-managed StackSet targeting the Workloads OU with automatic deployment.
- B. Use an SCP or appropriate resource policy to restrict deletion of the baseline resources.
- C. Ask each account administrator to copy the role manually.
- D. Use a tag policy and assume it creates the role.
- E. Use a Config aggregator as the deployment mechanism.
Show answer and reason
Answer: A and B.
Reason: Service-managed StackSets deploy the baseline to current and future accounts. A suitable preventive guardrail limits deletion; a Config aggregator provides visibility but does not deploy resources.
Question 2
An auditor asks for AWS’s SOC and ISO reports, while the security team needs evidence that its own controls operated during the audit period. Which services should be used?
- A. AWS Artifact for AWS reports and AWS Audit Manager for organizational evidence.
- B. Security Hub for both types of evidence.
- C. CloudFront and WAF logs for both types of evidence.
- D. IAM Access Advisor for AWS compliance reports.
Show answer and reason
Answer: A.
Reason: Artifact provides AWS’s compliance reports and agreements. Audit Manager collects evidence about the customer’s own controls and audit framework.
Question 3
A security team must centrally enforce WAF policies on CloudFront distributions and regional ALBs across organization accounts. Which managed service is designed for this organization-wide policy?
- A. AWS Firewall Manager with the appropriate WAF policies and scope.
- B. A separate manually maintained Lambda function in every account.
- C. An S3 lifecycle policy in the Log Archive account.
- D. A CloudFormation drift report with no remediation.
Show answer and reason
Answer: A.
Reason: Firewall Manager centrally applies supported firewall and WAF policies across accounts and resources. The policy scope and resource types still need to be configured correctly.
The Answer That Almost Works
These were the distractor rules I reviewed before the exam:
| Almost-Right Statement | Missing Detail |
|---|---|
| “The bucket policy allows the read.” | SSE-KMS also requires the correct caller and key-policy permissions. |
| “A security group can block this IP.” | Security groups allow traffic only; use a NACL or WAF when explicit deny behavior is required. |
| “Config detects the problem.” | Detection does not prevent creation or automatically remediate without a configured workflow. |
| “Security Hub sends directly to SNS.” | Use EventBridge to route findings to notifications or automation. |
| “Delete the compromised IAM user.” | Deactivate credentials and preserve identity and audit evidence first. |
| “Direct Connect encrypts traffic.” | Add a VPN when the requirement explicitly needs IPsec encryption. |
| “Automatic KMS rotation creates a new key every month.” | Automatic rotation keeps the KMS key ID; a new ID needs a key-transition design. |
| “The default S3 retention protects old objects.” | Existing object versions may need explicit retention treatment. |
| “A multi-account aggregator manages every account.” | Aggregation provides visibility; delegated administration or StackSets provide management or deployment. |
The service name was often correct but the boundary was wrong. That was the pattern I tried hardest to avoid.
Final Review Strategy: Read for the Security Requirement
At 170 minutes for 65 questions, the average budget is about 2 minutes 37 seconds per question.
My final reading order was:
- Identify the asset and owner.
- Underline the requirement verb: prevent, detect, contain, investigate, preserve, or prove.
- Trace the principal, policy, endpoint, route, service, resource, and key.
- Check whether the solution is cross-account, cross-Region, or organization-wide.
- Eliminate answers that use a real service outside its supported boundary.
- For multiple response, verify that every selected option is required and complete.
The question I kept asking
Resources and Further Reading
- SCS-C03 exam guide for the current exam domains and requirements.
- AWS Security Reference Architecture for multi-account security patterns.
- AWS Well-Architected Security Pillar for security design principles.
- IAM policy evaluation logic for identity and resource policy decisions.
- AWS KMS key policies, S3 Object Lock, and Amazon Security Lake for the boundaries discussed above.
- My SAP-C02 study notes for the architecture and scenario-review structure that I reused here.
Five Scenario Questions for Final Review
These questions combine multiple domains. Choose the complete architecture, not merely the answer containing the most familiar security service.
Scenario 1: Centralized Findings and Account Baselines
A company expects new accounts to join the Workloads OU. Every account needs a baseline IAM role, and the Security account needs centralized findings from GuardDuty and Inspector. Which combination minimizes custom onboarding? (Select TWO)
- A. Service-managed StackSet with automatic deployment to the Workloads OU.
- B. Manual baseline creation by each account owner.
- C. Delegated Security account administration for supported security services.
- D. A tag policy that creates the IAM role.
- E. Store all findings in the Organizations management account.
Show answer and reason
Answer: A and C.
Explanation: StackSets deploy the baseline to current and future OU accounts. Delegated administration centralizes supported security services without turning the management account into the workload and security-evidence boundary.
Scenario 2: Complete Cross-Account S3 Read
A role in Account B reads SSE-KMS objects in Account A through an interface endpoint. Access must be limited to that endpoint. Which combination must be reviewed?
- A. Caller IAM policy, S3 bucket policy, KMS key policy, endpoint policy, and matching conditions.
- B. Only the S3 bucket policy.
- C. A security group attached to the KMS key.
- D. An SCP that grants all required permissions.
Show answer and reason
Answer: A.
Explanation: The request crosses identity, resource, endpoint, and encryption boundaries. An SCP cannot grant the missing permissions, and KMS keys do not have security groups.
Scenario 3: Compromised Auto Scaling Instance
GuardDuty identifies malware on an instance behind an ALB and managed by an Auto Scaling group. Responders need to stop traffic, preserve the EBS volumes, and avoid sending the instance back into service. What is the correct approach?
- A. Terminate the instance immediately.
- B. Isolate it, remove it from service or replacement workflows, capture evidence, and then eradicate.
- C. Reboot it and run a patch baseline.
- D. Delete the entire Auto Scaling group.
Show answer and reason
Answer: B.
Explanation: The response must contain the compromise and preserve evidence before destructive actions. Broad deletion is not a controlled incident-response procedure.
Scenario 4: Private Partner Service
An application must call a partner service privately. The VPCs have overlapping CIDRs, and the partner must not initiate connections to the application. What should be used?
- A. VPC peering.
- B. Transit Gateway with full route propagation.
- C. PrivateLink interface endpoint.
- D. Public IP addresses with a broad security group.
Show answer and reason
Answer: C.
Explanation: PrivateLink provides consumer-initiated, service-specific connectivity without general bidirectional routing and without requiring non-overlapping VPC CIDRs.
Scenario 5: Immutable Compliance Evidence
Financial reports already exist in a versioned S3 bucket. Selected existing versions and future reports must not be deleted or have retention shortened for seven years, including by root. What should be configured?
- A. Object Lock Compliance mode with retention applied to selected existing versions and future objects.
- B. Governance mode with administrator bypass.
- C. An SCP denying
s3:DeleteObject. - D. A lifecycle expiration rule.
Show answer and reason
Answer: A.
Explanation: Compliance mode provides the required WORM behavior. Existing versions need explicit retention treatment; versioning alone, Governance mode, and policy denies do not meet the stated root-user threat model.
Final Thoughts
Passing SCS-C03 is a milestone I am glad to add to this certification journey. The badge records the result, but the useful habit is asking:
What is being protected, who can change it, what evidence must survive, and how does the control behave when the environment grows?
My advice