SAP-C03 · Domain 2: Security, Compliance, and Governance · 25% of the exam
Task 2.4: Design architectures to meet compliance requirements.
Meeting rules from outside: data protection, sovereignty and residency controls, audit trails for regulated workloads and AI services, compliance evidence from AWS Artifact, and threat modeling that drives mitigations.
Study it
Data protection, sovereignty, residency and audit trails for regulated workloads
Compliance evidence with AWS Artifact and threat modeling
Lesson coming
Sample questions
Try each one before opening the answer. Every option is explained, with the AWS documentation page that proves it.
Question 1 · choose 1
A company runs an AWS Control Tower landing zone with eu-central-1 as its home Region and 120 enrolled accounts in several OUs. A new data residency rule says that workloads in these accounts may run only in eu-central-1 and eu-west-1, and the rule must be enforced, not just reported. Global services such as IAM and AWS Organizations must keep working, accounts that Account Factory creates later must be covered automatically, and the platform team does not want to write or maintain its own policy documents for this. Which solution meets these requirements?
AKeep eu-central-1 and eu-west-1 as the only governed Regions and turn on the Region deny control in the landing zone settings
BRemove every other Region from the landing zone's governed Regions so that Control Tower stops covering those Regions
CAttach an SCP to the organization root that denies requests whose aws:RequestedRegion is not one of the two, exempting global services
DFrom the management account, disable every Region except the two in the Region settings of each member account
Show the answer and why
AKeep eu-central-1 and eu-west-1 as the only governed Regions and turn on the Region deny control in the landing zone settings
Correct
The landing zone Region deny control applies to every OU with the Control Tower baseline and is inherited by lower OUs. It denies requests in Regions that are not governed, keeps actions allowed in governed Regions, and exempts certain global services such as IAM and Organizations. Control Tower maintains the SCP behind it.
BRemove every other Region from the landing zone's governed Regions so that Control Tower stops covering those Regions
Incorrect
Governed Regions decide where Control Tower baselines accounts and applies its controls. A Region that is not governed can still be used; its resources are simply outside Control Tower governance, so nothing is enforced.
CAttach an SCP to the organization root that denies requests whose aws:RequestedRegion is not one of the two, exempting global services
Incorrect
A policy built on aws:RequestedRegion with a NotAction list of global services does enforce Region limits, but the team would write it and keep its list of exempted global services current itself, which it does not want to do.
DFrom the management account, disable every Region except the two in the Region settings of each member account
Incorrect
Only opt-in Regions can be enabled or disabled. Regions launched before March 20, 2019 are enabled by default and cannot be disabled, so most Regions would stay usable.
Three constraints decide it: enforcement rather than reporting, coverage of future accounts, and no custom policy to maintain. Narrowing the governed Regions alone only changes where Control Tower governs, not where people can deploy. Disabling Regions works only for opt-in Regions. A hand-written SCP enforces the rule but leaves the team maintaining it. The landing zone Region deny control enforces the governed Regions in every enrolled account, current and future, with global services exempted and the policy managed by Control Tower.
A new payments service must process card data on EC2 in an environment that even administrators of the parent instance cannot access, with no persistent storage and no interactive access. Which feature fits?
AA separate VPC for the payments service
BAWS Nitro Enclaves created from the EC2 instance
CA dedicated EC2 instance with SSH disabled and no key pairs
DAn encrypted EBS volume on the instance
Show the answer and why
AA separate VPC for the payments service
Incorrect
Network isolation does not stop administrators of the instance.
BAWS Nitro Enclaves created from the EC2 instance
Correct
Nitro Enclaves are isolated, hardened and highly constrained virtual machines created from EC2 instances.
CA dedicated EC2 instance with SSH disabled and no key pairs
Incorrect
Administrators of the instance could still change its configuration and access its memory.
DAn encrypted EBS volume on the instance
Incorrect
Encryption at rest does not isolate processing from administrators.
Isolated processing on EC2 is done in Nitro Enclaves.
A payments company's compliance team downloads AWS's PCI DSS attestation and SOC reports from AWS Artifact for its yearly audit. Last year the team missed a newly published report version and gave auditors an outdated copy. The team wants an email whenever reports or agreements it relies on are updated, without building anything. Which approach meets this requirement?
ACreate an AWS Config rule that checks the Artifact reports daily and notifies the team through Amazon SNS
BSchedule a Lambda function that lists Artifact reports through the API, compares versions and emails the team about every change it finds
CConfigure AWS Artifact email notifications for report and agreement updates, after selecting notification hubs in AWS User Notifications
DSubscribe the team to the AWS Health Dashboard, which posts an event when a new compliance report is available
Show the answer and why
ACreate an AWS Config rule that checks the Artifact reports daily and notifies the team through Amazon SNS
Incorrect
AWS Config evaluates the configuration of resources in the company's accounts. Reports published in AWS Artifact are not resources it records.
BSchedule a Lambda function that lists Artifact reports through the API, compares versions and emails the team about every change it finds
Incorrect
This would work but means building and running code, while AWS Artifact already sends notifications about report and agreement updates.
CConfigure AWS Artifact email notifications for report and agreement updates, after selecting notification hubs in AWS User Notifications
Correct
AWS Artifact can send email notifications about updates to agreements and reports through AWS User Notifications, once notification hubs are selected and a notification configuration names the recipients.
DSubscribe the team to the AWS Health Dashboard, which posts an event when a new compliance report is available
Incorrect
AWS Health reports events that affect AWS services and the company's resources. Report updates in AWS Artifact are delivered through Artifact notifications.
AWS Artifact is the self-service source of AWS compliance reports and agreements, and its email notifications, delivered through AWS User Notifications, tell the compliance team when those documents change.
An insurer in eu-west-1 calls a model in Amazon Bedrock through the EU geographic cross-Region inference profile. Its regulator asks for an audit record that shows, for each inference request, the Region in which the request was actually processed. The insurer already has an organization trail in AWS CloudTrail. Where can the insurer find this information?
AIn the CloudTrail events of the destination Regions, because each request is logged only where it was processed
BIn Amazon CloudWatch metrics for the inference profile, which publish one metric stream per destination Region
CIn VPC Flow Logs of the application's subnets, which show the IP addresses of the Bedrock endpoints that answered
DIn the CloudTrail event for each request in the source Region, in the additionalEventData.inferenceRegion field
Show the answer and why
AIn the CloudTrail events of the destination Regions, because each request is logged only where it was processed
Incorrect
CloudTrail logs every cross-Region inference request in the source Region, where the application made the call, not only in the Region that processed it.
BIn Amazon CloudWatch metrics for the inference profile, which publish one metric stream per destination Region
Incorrect
Metrics show aggregate volumes and latencies. They do not give a per-request audit record of where each request ran.
CIn VPC Flow Logs of the application's subnets, which show the IP addresses of the Bedrock endpoints that answered
Incorrect
The application connects to the Bedrock endpoint in its own Region. Flow logs cannot show where Bedrock routed the request after that.
DIn the CloudTrail event for each request in the source Region, in the additionalEventData.inferenceRegion field
Correct
CloudTrail records cross-Region inference requests in the source Region, and the additionalEventData.inferenceRegion field identifies the Region where each request was processed.
With cross-Region inference, the audit trail stays in the source Region's CloudTrail, and each event names the processing Region, which supports data residency evidence for regulated AI workloads.
A European bank's data residency policy allows customer data to be processed only in EU Regions. An SCP on its workload OU denies all actions outside eu-central-1 and eu-west-1. A chatbot in eu-central-1 moved from an in-Region model ID to the EU geographic cross-Region inference profile for more throughput, and its role allows the profile and the model in every destination Region. Since then, a share of requests fails with AccessDenied errors. The bank must keep both the residency policy and the extra throughput. What should a solutions architect do?
ASwitch the chatbot to the global cross-Region inference profile, which spreads requests across more Regions
BMove the chatbot back to the in-Region model ID in eu-central-1 and request higher on-demand quotas there
CDetach the Region-deny SCP from the workload OU and rely on the geographic inference profile alone to keep every request inside the EU
DAdd the EU destination Regions of the eu-central-1 inference profile to the SCP's allowed Regions
Show the answer and why
ASwitch the chatbot to the global cross-Region inference profile, which spreads requests across more Regions
Incorrect
Global profiles route requests to any supported commercial Region worldwide, which breaks the EU-only residency policy.
BMove the chatbot back to the in-Region model ID in eu-central-1 and request higher on-demand quotas there
Incorrect
In-Region invocation avoids the SCP problem, but the bank wants to keep the extra throughput that the cross-Region profile provides.
CDetach the Region-deny SCP from the workload OU and rely on the geographic inference profile alone to keep every request inside the EU
Incorrect
Removing the SCP drops the organization-wide control that enforces the residency policy for every other service in the OU.
DAdd the EU destination Regions of the eu-central-1 inference profile to the SCP's allowed Regions
Correct
An SCP that blocks a destination Region makes cross-Region inference fail when a request is routed there. Allowing every EU destination Region of the profile keeps processing in the EU and restores the routing.
Geographic cross-Region inference keeps data in the chosen geography, but a Region-allowlist SCP must allow every destination Region listed in the profile for the source Region, or include an inference-profile exception.
A regulated insurer created a multi-Region organization trail in AWS CloudTrail from its management account, with Europe (Milan) as the trail's home Region, delivering to a bucket in its log archive account. Milan is an opt-in Region, and most of the 80 member accounts never enabled it. Auditors found that the trail holds no events from those accounts. The audit trail must cover every account, including accounts created later, in all the Regions they use, and administrators of member accounts must not be able to stop or change the logging. What should a solutions architect do?
ARecreate the multi-Region organization trail in a Region that is enabled by default, delivering to the same bucket
BKeep the trail, and deploy a multi-Region trail with StackSets into each member account that has not enabled Milan
CRegister the log archive account as a CloudTrail delegated administrator and create the trail from there in Milan
DReplace the trail with a single-Region organization trail whose home Region is enabled by default in every account
Show the answer and why
ARecreate the multi-Region organization trail in a Region that is enabled by default, delivering to the same bucket
Correct
When a multi-Region organization trail's home Region is an opt-in Region, member accounts send no activity unless they opt in to that Region. With a default Region as home, the trail logs every enabled Region of every member account, and members cannot change it.
BKeep the trail, and deploy a multi-Region trail with StackSets into each member account that has not enabled Milan
Incorrect
Account trails would close the coverage gap, but each belongs to its member account, so that account's administrators could stop or delete it, which member accounts cannot do to an organization trail.
CRegister the log archive account as a CloudTrail delegated administrator and create the trail from there in Milan
Incorrect
A delegated administrator can manage organization trails, which members cannot change, but the opt-in home Region rule still applies, so the same accounts would send no events.
DReplace the trail with a single-Region organization trail whose home Region is enabled by default in every account
Incorrect
A single-Region organization trail cannot be changed by members, but it logs activity only in its home Region, not in the other Regions the accounts use.
Organization trails protect the audit trail from member accounts, and the home Region decides who can send to them: a multi-Region organization trail homed in an opt-in Region receives activity only from accounts that enabled that Region.
A health insurer is designing a claims platform that will process protected health information with API Gateway, Lambda, DynamoDB and Amazon Bedrock. Before any code is written, the risk committee wants the threats to this design identified and tied to mitigations, and it wants the review effort spent where the insurer can actually change something. Which actions follow AWS guidance for this risk assessment? (Choose TWO.)
ATurn on Amazon GuardDuty in the platform accounts and review its findings with the risk committee every week
BWalk the design's data flow diagram with the STRIDE categories and record each threat with its mitigation as a threat statement
CSchedule a penetration test of the platform once it is live, and fix the findings in the next release
DAsk for each threat who is responsible, and scope the mitigations to the insurer's side, such as service configuration and custom code
EAdopt the OWASP Top 10 list as the platform's threat model and assign one mitigation to each of its entries
Show the answer and why
ATurn on Amazon GuardDuty in the platform accounts and review its findings with the risk committee every week
Incorrect
GuardDuty detects threats in accounts that already run workloads. It cannot assess a design that has not been built.
BWalk the design's data flow diagram with the STRIDE categories and record each threat with its mitigation as a threat statement
Correct
A threat model starts from a model of the system, such as a data flow diagram, and a framework such as STRIDE helps the team enumerate what can go wrong; tools such as Threat Composer record threat statements and mitigations.
CSchedule a penetration test of the platform once it is live, and fix the findings in the next release
Incorrect
A penetration test needs a running system, so it comes after the code exists and cannot drive design decisions before development.
DAsk for each threat who is responsible, and scope the mitigations to the insurer's side, such as service configuration and custom code
Correct
AWS guidance pairs "What are we going to do about it?" with "Who is responsible?": the shared responsibility model scopes mitigations to the controls under the customer's control, typically service configuration and the company's own code.
EAdopt the OWASP Top 10 list as the platform's threat model and assign one mitigation to each of its entries
Incorrect
Published lists such as the OWASP Top 10 can inspire brainstorming, but there is no canonical list of what can go wrong: the threats come from the team's own model of this system.
Threat modeling works best early in design and stays a living document. A data flow diagram and STRIDE structure the threats, and the shared responsibility question keeps mitigations on what the customer controls.
Employees authenticate with the corporate identity provider and use IAM SAML 2.0 federation (AssumeRoleWithSAML) to assume one shared IAM role that can read sensitive files in an S3 bucket. The identity provider sets the role session name to the employee's department, and other teams' reports depend on those session names, so they must stay as they are. An audit found that the company cannot tell which person read a given object: the account's CloudTrail trail records only management events. Auditors want each object-read record to name the person directly, without joining it to sign-in events. Which changes meet these requirements? (Choose TWO.)
ATurn on VPC Flow Logs for every VPC from which employees reach Amazon S3 and query the records with Athena
BAdd S3 data events for the sensitive bucket to the trail with an advanced event selector on its ARN
CHave the identity provider send the user ID as SAML SourceIdentity and allow sts:SetSourceIdentity in the role trust policy
DTurn on Amazon GuardDuty S3 Protection in the account so that GuardDuty analyzes every object read in the bucket
ERecord the bucket in AWS Config and review its configuration timeline to find each read of an object
Show the answer and why
ATurn on VPC Flow Logs for every VPC from which employees reach Amazon S3 and query the records with Athena
Incorrect
Flow logs capture information about the IP traffic going to and from network interfaces. They do not show which object was read or which person sent the request.
BAdd S3 data events for the sensitive bucket to the trail with an advanced event selector on its ARN
Correct
Trails do not log data events by default. Object-level operations such as GetObject are S3 data events, and an advanced event selector can limit them to the objects of one bucket.
CHave the identity provider send the user ID as SAML SourceIdentity and allow sts:SetSourceIdentity in the role trust policy
Correct
The SourceIdentity attribute in the SAML assertion sets a value that stays with every request of the role session and cannot be changed during it, and CloudTrail shows it in the userIdentity of the events from that session. The role trust policy must allow sts:SetSourceIdentity, or the role assumption fails.
DTurn on Amazon GuardDuty S3 Protection in the account so that GuardDuty analyzes every object read in the bucket
Incorrect
S3 Protection analyzes S3 data events without needing data event logging on a trail, and it produces findings for potential threats. It does not give the company a record of every read.
ERecord the bucket in AWS Config and review its configuration timeline to find each read of an object
Incorrect
AWS Config records how resources are configured and how those configurations change over time, not reads of the objects in a bucket.
Two gaps need closing. Object reads are data events, which a trail logs only after they are added to it. And because the session names on the shared role do not identify anyone and must stay as they are, the identity provider passes the person's ID as the source identity in the SAML assertion; CloudTrail then shows that value on each action taken in the session, including each GetObject data event.