www.awsgood.com
Threat Detection and Incident Response reads like the smallest slice of the blueprint, and plenty of people treat it as a rounding error. That's a mistake, because Domain 1 is the only place the exam asks you to reason under time pressure about a live compromise. The other five domains ask "is this configured correctly"; this one asks "the finding fired at 03:00, what do you do in the next ten minutes."
| Domain | Weight | What it actually tests |
|---|---|---|
| 1. Threat Detection and Incident Response | 14% | GuardDuty, Security Hub, Detective, EventBridge automation, IR lifecycle |
| 2. Security Logging and Monitoring | 18% | CloudTrail, VPC Flow Logs, Config, log integrity, Athena queries |
| 3. Infrastructure Security | 20% | Edge protection, VPC design, network segmentation, EC2/EKS hardening |
| 4. Identity and Access Management | 16% | Permission boundaries, SCPs, role trust policy abuse, federation |
| 5. Data Protection | 18% | KMS key policies, S3 encryption, Macie, secrets management |
| 6. Management and Security Governance | 14% | Organizations, Control Tower, Security Hub aggregation, compliance |
Sixty-five questions times 14% is about nine items. Expect two or three of those to be long scenario stems with four plausible remediation options, which is where most of your clock disappears.
| Item | Detail |
|---|---|
| Questions / time | 65 questions, 170 minutes (~2.6 min each) |
| Passing score | 750 on a 100–1000 scaled range |
| Fee | $300 USD list; local pricing and tax vary by region |
| Delivery | Pearson VUE test center or online proctored |
| Validity | 3 years |
| Prerequisite | None officially — hands-on experience assumed |
The question nobody asks first is whether SCS-C03 should be your first AWS exam. For most SMB ops engineers it shouldn't be. SAA-C03 gives you the vocabulary of VPCs, IAM roles and KMS that the security exam assumes you already own, and the overlap is real: about a third of the security scenarios are architect questions with a threat model bolted on.
If you have operated GuardDuty, Security Hub and an organization trail in a real account for six months or more, jump straight in — the exam validates things you already do. If your security experience is "I turned on default encryption on S3," start with SAA-C03 and come back.
Don't book it if your only exposure is console clicking in a sandbox you touched twice. The exam assumes you know what AccessDenied looks like when a KMS key policy is missing kms:Decrypt for the CloudTrail role, and why a trail that stops writing is worse than a trail that never existed.
Skip it too if you need security depth for exactly one workload and nothing else — reading the Well-Architected Security Pillar and the GuardDuty finding-types reference will get you further for free. And don't book 48 hours before a product launch; the reschedule window is 24 hours and the cognitive load isn't worth it.
| Week | Focus | Drill you actually run |
|---|---|---|
| 1 | CloudTrail + Config + VPC Flow Logs | Read one week of your own event history with Athena |
| 2 | GuardDuty + Security Hub (Domain 1 core) | Generate a sample finding, route it to Slack |
| 3 | IAM, SCPs, permission boundaries | Break a role's trust policy and fix it |
| 4 | KMS, S3, Macie, Secrets Manager | Rotate a key, then find every consumer that broke |
| 5 | Network security, Detective, Inspector | Recreate a VPC with segmented subnets |
| 6 | Practice exam + weak-area review | Timed 65-question run, then re-read misses |
The drills matter more than the reading. Two commands I run in every new account before I trust any detection story I tell an interviewer or an examiner:
# What detectors exist, and are the S3/K8s protections actually on?
aws guardduty list-detectors --region us-east-1
aws guardduty get-detector --detector-id 12abc34d5ef6789abcdef01EXAMPLE --region us-east-1
--query '{S3:DataSources.S3Logs.Status,K8s:DataSources.Kubernetes.AuditLogs.Status,Features:Features[*].[Name,Status]}'
The output is usually a reminder that GuardDuty's S3 protection defaults never got enabled on that one legacy account — which is exactly the kind of gap a Domain 1 scenario will ask you to notice.
For the automated-remediation half of the domain, wire a real rule and look at what it produced:
```bash
cat > gd-pattern.json <<'JSON'
{
"source": ["aws.guardduty"],
"detail-type": ["GuardDuty Finding"],
"detail": { "severity": [{ "numeric": [">=", 7] }] }
}
JSON
aws events put-rule --name gd-high-severity
--event-pattern file://gd-pattern.json --region us-east-1
Severity 7 and above is the line the exam treats as "wake someone up." Below that, a dashboard is fine.
## Field note: the 03:12 finding that never paged anyone
I had this setup working in a lab and then watched it fail in a real account. On 2026-11-03 at 03:12 UTC, GuardDuty raised `UnauthorizedAccess:IAMUser/InstanceCredentialExfiltration.OutsideAWS` against an m5.xlarge running a batch job. The finding was real: credentials from the instance profile were being used from an IP outside AWS.
Nothing paged. The on-call spotted it 41 minutes later in the console, which for that finding type is far too long.
The EventBridge rule had matched — the invocation happened, the target was invoked. What broke was delivery to SNS, because the topic policy only listed the account root and `cloudwatch.amazonaws.com`. EventBridge publishes to SNS as `events.amazonaws.com`, and when permission is missing the failure is quiet from the rule's perspective; it's a metric, not an error page.
I confirmed it with two numbers instead of guessing:
```bash
aws cloudwatch get-metric-statistics --namespace AWS/Events
--metric-name FailedInvocations
--dimensions Name=RuleName,Value=gd-high-severity
--start-time 2026-11-03T03:00:00Z --end-time 2026-11-03T05:00:00Z
--period 300 --statistics Sum --region us-east-1
`Sum: 2.0` across the window, plus a matching `NumberOfNotificationsFailed` on the topic itself. That pair is the fingerprint of a resource-policy problem, not a rule-pattern problem.
The fix was a topic policy statement scoped to the rule ARN:
```json
{
"Sid": "AllowEventBridgePublish",
"Effect": "Allow",
"Principal": { "Service": "events.amazonaws.com" },
"Action": "sns:Publish",
"Resource": "arn:aws:sns:us-east-1:111122223333:sec-alerts",
"Condition": {
"ArnLike": { "aws:SourceArn": "arn:aws:events:us-east-1:111122223333:rule/gd-high-severity" }
}
}
Verified by publishing a test event through `aws events put-events` with `source: aws.guardduty` and watching the SNS delivery metric increment instead of the failure metric. The lesson that stuck: a detection pipeline is only as good as its delivery path, and the exam loves that exact distinction. If a question asks why findings aren't reaching the SOC, check the topic policy before the rule.
## Trap answers Domain 1 questions are built around
These are the patterns I'd bet on seeing:
- **"Terminate the instance to stop the attack"** — wrong when evidence preservation is on the table. Isolate (security group, or the GuardDuty-initiated isolation) and snapshot first.
- **"Delete the access key"** for an active exfiltration — deleting doesn't invalidate already-issued temporary credentials. The correct move is an explicit deny on the principal with an `aws:TokenIssueTime` condition.
- **Detective vs. Macie vs. Inspector confusion.** "What happened and what else is related" is Detective. "Is this bucket holding PII" is Macie. "Which EC2 instance has an unpatched CVE" is Inspector.
- **Security Hub aggregation across regions** requires a delegated administrator in the management account — not an invitation.
- **Runbook order.** Containment before eradication, always. NIST 800-61 ordering shows up verbatim in the distractors.
Notice how many of these are about ordering and scope rather than product trivia. My advice: when you're stuck between two options, pick the one that limits blast radius without destroying forensic state.
## Exam technique under a 170-minute clock
Two passes. First pass answers everything you can settle in under 90 seconds and flags the rest; second pass spends the remaining time on scenarios. Set a hard checkpoint: 33 questions by minute 70, or you'll run out of room on the last five.
Read the final sentence of the stem first — it usually contains the constraint ("LEAST operational overhead", "MOST secure", "with the fewest changes") and that constraint eliminates two answers before you've read the paragraphs. Multi-response questions tell you how many to pick; don't second-guess it.
If you run into a question about a service you've never touched, look for the option that uses a managed AWS service with a service principal rather than a custom Lambda doing credential work. That heuristic is right often enough to matter.
For practice, the official AWS practice exam plus a timed run from any third-party bank is enough — do the timed run at least once, because the 170-minute clock is part of the test. When our team runs through these certs internally, we usually pair the study plan with a light awsgood.com managed-services review of the account's GuardDuty and CloudTrail setup, so the lab work happens against something real.
## FAQ
**Do I need GuardDuty in production to pass?** No, but you need to have seen its output. A sandbox detector with sample findings enabled is enough to understand severity, finding types and the isolation workflow.
**Is the exam multiple choice only?** Mostly multiple choice with some multiple-response items where you select two or three correct answers. There is no hands-on lab component.
**How long is the SCS-C03 study window for someone working full time?** Six weeks at 7–8 hours a week, or about 40–50 hours total if you already operate the logging and detection stack. From scratch, double it.
## Sources
- [AWS Certified Security – Specialty certification page](https://aws.amazon.com/certification/certified-security-specialty/)
- [SCS-C03 Exam Guide (AWS Certification docs)](https://docs.aws.amazon.com/aws-certification/latest/security-specialty-03/security-specialty-03.html)
- [GuardDuty IAM finding types reference](https://docs.aws.amazon.com/guardduty/latest/ug/guardduty_finding-types-iam.html)
- [Using resource-based policies with Amazon EventBridge](https://docs.aws.amazon.com/eventbridge/latest/userguide/eb-use-resource-based.html)