AWS Certified Security – Specialty (SCS-C03)
The most important exam for banking. It covers detection, incident response, infrastructure security, IAM, data protection and governance.
Verified against official sources on 2026-11-17.
AWS Certified Security – Specialty (SCS-C03) checks that you can secure AWS workloads. That means classifying and encrypting data, designing identity at scale, building detection and incident response, securing networks, and governing many accounts. For a banking developer, it's the certification that best matches what regulators and security teams expect from you.
At a glance
| Code | SCS-C03 (replaced SCS-C02 on 2025-12-02) |
| Level | Specialty |
| Intended audience | The equivalent of 3–5 years securing cloud solutions |
| Questions | 65 (50 scored + 15 unscored) |
| Time | 170 minutes |
| Passing score | 750 (scaled 100–1000) |
| Price | USD 300 |
| Question types | Multiple choice, multiple response, ordering (put 3–5 steps in order), matching (match 3–7 prompts) |
| Valid for | 3 years. Renew by passing the latest SCS exam, or with the paid Skill Builder maintain option (adds 1 year) |
| Out of scope | Designing cryptographic algorithms, packet-level traffic analysis, overall cloud architecture, end-user compute, ML training |
Official domains and weightings
| Domain | Weight |
|---|---|
| 1. Detection | 16% |
| 2. Incident Response | 14% |
| 3. Infrastructure Security | 18% |
| 4. Identity and Access Management | 20% |
| 5. Data Protection | 18% |
| 6. Security Foundations and Governance | 14% |
What's new compared with SCS-C02: the old "Threat Detection and Incident Response" and "Logging and Monitoring" domains have become Detection and Incident Response. IAM grew from 16% to 20%. New topics include guardrails for GenAI applications (OWASP Top 10 for LLM applications), ingesting OCSF-format data, in-transit encryption between nodes, imported versus AWS-generated key material, masking sensitive data (CloudWatch Logs data protection, SNS message data protection), and multi-Region keys and certificates. Basic TCP/IP, TLS concepts and S3 static website hosting were removed.
Key AWS services tested
- Detection: CloudTrail (organization trails, log file validation), GuardDuty, Security Hub, Amazon Detective, Amazon Security Lake, Config, CloudWatch, VPC Flow Logs, Inspector, Macie
- Response: EventBridge, Lambda and Step Functions for automated response, Systems Manager (Automation, Run Command), EBS snapshots for forensics, isolation security groups
- Infrastructure: VPC security groups and NACLs, AWS WAF, Shield Advanced, Network Firewall, Route 53 Resolver DNS Firewall, CloudFront, PrivateLink, Firewall Manager
- Identity: IAM policy evaluation, permission boundaries, SCPs and RCPs, IAM Identity Center, IAM Access Analyzer, STS and role chaining, Cognito, Verified Permissions, IAM Roles Anywhere
- Data protection: KMS (key policies, grants, multi-Region keys, imported key material), CloudHSM, Secrets Manager, ACM and Private CA, S3 Object Lock, Macie, Backup Vault Lock
- Governance: Organizations, Control Tower, Config conformance packs, Audit Manager, AWS Artifact, CloudFormation Guard, Service Catalog
What you already know vs what you must learn
| You already know (from .NET and banking) | You must learn |
|---|---|
| OWASP Top 10, secure coding, parameterised queries | The IAM policy evaluation order: explicit deny, then SCP and RCP, resource policy, permission boundary, session policy, identity policy |
| TLS, certificates, maybe SQL Server TDE or Always Encrypted | KMS: key policies versus IAM policies, grants, envelope encryption, key rotation, cross-account access |
| Segregation of duties, audit logging, change control | Multi-account guardrails: SCPs, RCPs, Control Tower controls, delegated administrator accounts |
| How your bank's SOC and incident process works | AWS incident response mechanics: isolating an EC2 instance, capturing snapshots, rotating credentials, GuardDuty finding types |
| PCI DSS and GDPR concepts | Which AWS service produces which evidence, and where AWS's responsibility ends |
Preparation order and study plan (about 60–80 hours over 6–8 weeks, after SAA and DVA)
| Week | Focus | Hours |
|---|---|---|
| 1 | Exam guide and the C02→C03 comparison appendix. Module 05 identity section (Domain 4) | 10 |
| 2 | IAM deep dive: policy evaluation exercises, SCPs, RCPs, permission boundaries, Access Analyzer | 10 |
| 3 | Data protection (Domain 5): KMS labs, Secrets Manager rotation, Object Lock, Macie, log masking | 10 |
| 4 | Detection (Domain 1): organization trail, GuardDuty, Security Hub, Config. Module 11 | 8–10 |
| 5 | Incident response (Domain 2): runbooks, isolation, forensics. Module 17 playbooks | 8 |
| 6 | Infrastructure (Domain 3): WAF, Network Firewall, endpoints. Governance (Domain 6): Module 15 | 8–10 |
| 7–8 | Free Exam Prep course, the 20-question set, ordering and matching practice, mock exams | 10–12 |
Mapped platform modules
- Module 05 — Security for banking: the core of every domain
- Module 02 — Networking: segmentation, endpoints, Network Firewall
- Module 11 — Monitoring: CloudTrail, CloudWatch, logging strategy
- Module 15 — Governance & FinOps: Organizations, Control Tower, Config, Audit Manager
- Module 17 — Playbooks:
AccessDenied, KMS errors, public buckets, "who accessed this record?" - Module 16 — GenAI: GenAI guardrails (new Task 3.2.7)
Practical skills to build
- Set up an organization trail into a locked-down log-archive account with log file validation, and query it with Athena
- Write and test an SCP that denies disabling CloudTrail and GuardDuty and restricts which Regions can be used
- Share a customer managed KMS key with another account (key policy plus IAM policy) and explain each statement
- Add a CloudWatch Logs data protection policy that masks card numbers in API logs
- Run a GuardDuty sample finding through EventBridge to a Lambda function that isolates the instance and snapshots its volumes
- Use IAM Access Analyzer to find resources shared outside your organization
Exam-prep checklist
- I can work through IAM policy evaluation in any order AWS presents it, including cross-account cases
- I know when a KMS key policy alone is enough and when an IAM policy is also needed
- I know how SCPs, RCPs and permission boundaries differ (what each one limits and what it never grants)
- I can list the incident response steps for a compromised EC2 instance and a compromised access key, in order
- I know which logs answer which question: CloudTrail, VPC Flow Logs, S3 server access logs, ALB logs, DNS query logs
- I know how WAF, Shield Advanced, Network Firewall and security groups each apply to a scenario
- I can explain Object Lock governance mode versus compliance mode, and Backup Vault Lock
- I've practised ordering and matching questions
Official links
- Exam guide (SCS-C03)
- Comparison of SCS-C02 and SCS-C03
- Exam product page
- Skill Builder Exam Prep plan (SCS-C03)
Practice questions
These 5 questions are original, banking-flavoured scenarios. They aren't real exam questions.
A data account holds transaction files in S3, encrypted with SSE-KMS using a customer managed key. An analytics account's role needs to read and decrypt them. The bucket policy already allows the analytics role to call s3:GetObject. Reads still fail with AccessDenied from KMS. What must be done? (Choose the most complete answer.)
- A. Re-encrypt the objects with the AWS managed key
aws/s3and share that key - B. Add a statement to the key policy that allows the analytics account (or its role)
kms:Decrypt, and attach an IAM policy to the analytics role that allowskms:Decrypton that key's ARN - C. Create a KMS grant from the analytics account to itself
- D. Export the key material and import it into the analytics account
Correct answer: B
Explanation: Cross-account KMS access needs permission on both sides. The key's owner must trust the external principal in the key policy, and the external account must allow its role to use the key in an IAM policy. This two-sided model is deliberate: neither account can give itself access to the other's key on its own. To narrow access further, add a kms:ViaService condition so the key can only be used through S3.
Why the others are wrong:
- A: You can't edit the key policies of AWS managed keys, so they can't be shared across accounts.
- C: Only principals with permission on the key, such as its owner, can create grants. The analytics account has no such permission.
- D: You can't export key material from KMS, and copying keys would destroy the separation of control.
GuardDuty reports that an EC2 instance running a batch job is talking to a known cryptocurrency-mining domain. The bank's incident response policy requires containment while preserving evidence for forensics. What should the responder do first?
- A. Terminate the instance immediately to stop the activity
- B. Isolate the instance by replacing its security groups with an isolation security group, detach it from its Auto Scaling group and load balancer, take EBS snapshots and tag the instance, then investigate (and capture memory if your process requires it)
- C. Reboot the instance to clear the malicious process
- D. Archive the GuardDuty finding and open a ticket for next week
Correct answer: B
Explanation: Contain without destroying evidence. Isolate the instance on the network, stop the Auto Scaling group from replacing or terminating it, snapshot the volumes for offline analysis, and tag it so that everyone knows it's under investigation. One subtlety: changing security group rules doesn't always cut off connections that are already tracked. AWS's incident response guidance explains how to deal with this, for example by also using network ACLs. Then rotate any credentials the instance could reach, such as its instance role session.
Why the others are wrong:
- A: Terminating destroys volatile evidence (memory, running processes) and may delete the EBS volumes as well.
- C: Rebooting loses volatile evidence and doesn't contain anything.
- D: Delaying the response to an active compromise is unacceptable.
A bank's cloud platform team must make sure that no one in any member account, including local administrators, can stop CloudTrail logging or remove the account from the organization. What's the most effective control?
- A. An IAM policy attached to every administrator in each account
- B. An AWS Config rule that detects when CloudTrail is stopped
- C. A service control policy (SCP) on the organization's root or OUs that denies
cloudtrail:StopLogging,cloudtrail:DeleteTrailandorganizations:LeaveOrganization - D. A permission boundary on the administrator role
Correct answer: C
Explanation: SCPs set the maximum permissions for every principal in member accounts, including the account's root user. An explicit deny in an SCP can't be overridden by anything inside the account. That makes SCPs the right tool for organisation-wide preventive guardrails. Test SCPs in a sandbox OU first. SCPs don't affect the management account, which is one reason to keep workloads out of it.
Why the others are wrong:
- A: Local administrators could edit or detach the IAM policy.
- B: Config is detective: it tells you after the event. Use it alongside the SCP, not instead of it.
- D: Permission boundaries apply only to the principals they're attached to, and an administrator could create a new role without one.
A code review finds that a .NET payments API has been writing full card numbers (PANs) into CloudWatch Logs through a verbose exception logger. The fix will take two sprints. In the meantime, the bank needs any new log events to have card numbers masked for everyone except a small, approved investigation role. What should the team do?
- A. Use Amazon Macie on the log group
- B. Add a CloudWatch Logs data protection policy to the log group with the credit card number data identifier. Grant
logs:Unmaskonly to the investigation role - C. Encrypt the log group with a customer managed KMS key
- D. Shorten the log group's retention to 1 day
Correct answer: B
Explanation: CloudWatch Logs data protection policies detect sensitive data using managed data identifiers (financial ones included) and mask it in log events stored after the policy is enabled. Only principals with logs:Unmask can see the original values. This counts as compensating control, not a fix: the code must be corrected, and the bank's PCI DSS team must assess the exposure that already happened.
Why the others are wrong:
- A: Macie scans S3 objects, not CloudWatch Logs.
- C: KMS encryption protects data at rest. Anyone with read access to the logs still sees the card numbers.
- D: Shorter retention lowers the exposure window, but card numbers are still visible while the logs exist, and it may break audit retention rules.
A developer's IAM user access key, still used by a legacy on-premises batch job, was found in a public Git commit. Put these response steps in the correct order so that you contain the exposure without unplanned downtime (assuming an immediate outage isn't acceptable).
- Deactivate the old access key
- Create a second access key for the user
- Check the old key's "last used" information and CloudTrail for unauthorised use, and keep monitoring
- Update the batch job to use the new key and confirm that it works
- Delete the old access key
Correct order: 2 → 4 → 1 → 3 → 5
Explanation: An IAM user can have two access keys, which allows rotation without downtime. Create the new key, switch the workload to it, then deactivate the old one. Deactivation is reversible, so if something breaks you can turn it back on briefly. Investigate CloudTrail activity tied to the exposed key, then delete it. Because the key is public, every minute counts. If you see signs of misuse, deactivate it immediately and accept the outage.
Why other orders are wrong:
- Deactivating first (1 before 2 and 4) stops the batch job at once, which goes against the no-downtime requirement unless misuse is confirmed.
- Deleting before checking (5 before 3) loses the chance to undo if you missed a dependency. Investigation can still happen with CloudTrail afterwards, but it's safer to check first.
The real lesson: long-lived keys are the root problem. Replace them with IAM Roles Anywhere (for on-premises workloads) or with roles and temporary credentials, and add secret scanning to CI.
Independent learning resource — not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
AWS Certified Developer – Associate (DVA-C03)
The developer exam. It tests building, securing, testing, deploying and debugging applications on AWS, now including AI-assisted development.
AWS Certified Data Engineer – Associate (DEA-C01)
The data pipeline exam. It covers ingesting, transforming, storing, operating and governing data on AWS.