# Access Control

Access control determines which identities can perform which actions on which resources under specified conditions.

Cloud concept. Updated 2026-10-29. Page: https://blowstack.com/tools/cloud-concepts-hub/access-control

> Access control turns security requirements into enforceable permissions for people, applications, and cloud services. It limits unauthorized activity while preserving legitimate operations, including the ability to deliver logs, run health checks, and investigate incidents.

Practice exams covering Access Control: [ANS-C01 AWS Certified Advanced Networking - Specialty](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty), [AIB-C01 (beta) AWS Certified AI Business Strategist](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist), [AIF-C01 AWS Certified AI Practitioner](https://blowstack.com/practice-exams/aws-certified-ai-practitioner), [CLF-C02 AWS Certified Cloud Practitioner](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner), [DEA-C01 AWS Certified Data Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-data-engineer-associate), [DVA-C02 AWS Certified Developer - Associate](https://blowstack.com/practice-exams/aws-certified-developer-associate), [DOP-C02 AWS Certified DevOps Engineer - Professional](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional), [MLA-C02 (beta) AWS Certified Machine Learning Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate), [SCS-C03 AWS Certified Security - Specialty](https://blowstack.com/practice-exams/aws-certified-security-specialty), [SAA-C03 AWS Certified Solutions Architect - Associate](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate), [SAP-C03 AWS Certified Solutions Architect - Professional](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional).

## 6 key aspects of Access Control

Effective access control combines scoped permissions, layered enforcement, explicit service authorization, and evidence-driven validation without confusing infrastructure placement with security boundaries.

1. **Identity, action, resource, and context**: Evaluate the requester, requested operation, target resource, and relevant conditions together. Authentication proves identity; authorization determines whether that identity may perform the operation.
2. **Least privilege and temporary access**: Scope permissions to the task and prefer temporary workload or federated role credentials over embedded long-lived keys. Review unused permissions and provide controlled, auditable emergency access.
3. **Layered policy evaluation**: Consider identity policies, resource policies, organizational restrictions, and session limits when diagnosing access decisions. In AWS, an applicable explicit deny overrides an allow, and permission-limiting policies do not independently grant access.
4. **Separation of location and authorization**: Treat Regions, Availability Zones, and edge locations as infrastructure placement and delivery concepts, not automatic permission boundaries. Apply explicit controls when access must be constrained by geography or destination.
5. **Service-to-service permissions**: Give workloads and delivery services the permissions needed to invoke dependencies and publish telemetry. Check role trust, destination policies, and encryption-key access rather than assuming that enabling a feature completes its authorization setup.
6. **Observable and testable enforcement**: Test both permitted and denied operations, and preserve evidence of policy changes and access failures. Distinguish authorization failures from network, logging-configuration, and destination-selection problems.

## Cloud services for Access Control

IAM defines permissions for engineers, workloads, and service interactions. Lambda and API Gateway apply those permissions to execution and invocation paths, while CloudFront adds separate viewer and origin access controls at the delivery layer. CloudWatch receives operational telemetry, and CloudTrail records supported activity for auditing and investigation. These services require independent authorization checks: permission to invoke an application does not imply permission to publish its logs or read its audit records. Troubleshoot each producer-to-destination relationship, including role trust, resource policies, encryption permissions, and logging configuration.

- **[Amazon CloudWatch](https://blowstack.com/tools/cloud-services-hub/amazon-cloudwatch)**: CloudWatch provides logs, metrics, and alarms whose publishing, viewing, and administration require appropriate permissions. For log-delivery troubleshooting, inspect producer permissions, destination policies where applicable, and encryption settings separately from alarm configuration.
- **[AWS Lambda](https://blowstack.com/tools/cloud-services-hub/aws-lambda)**: Lambda separates a function's execution-role permissions from permissions that allow callers or services to invoke it. Missing execution-role permissions can prevent CloudWatch Logs delivery even when the function runs.
- **[Amazon API Gateway](https://blowstack.com/tools/cloud-services-hub/amazon-api-gateway)**: API Gateway supports request authorization through mechanisms that vary by API type, including IAM authorization and authorizers. Logging must also be configured with the permissions and destinations required by the selected API type and logging feature.
- **[Amazon CloudFront](https://blowstack.com/tools/cloud-services-hub/amazon-cloudfront)**: CloudFront can restrict viewer access with signed URLs or cookies and use origin access control for supported origins. Viewer authorization, origin authorization, and log-delivery permissions are separate configurations.
- **[AWS CloudTrail](https://blowstack.com/tools/cloud-services-hub/aws-cloudtrail)**: CloudTrail records supported account and API activity for access investigations and change auditing. Configure the necessary event coverage, including data events where required, and protect both log delivery and access to the resulting records.

## Access Control exam questions

Real practice exam questions on Access Control, with the correct answer and its explanation.

### Question 1 (AIB-C01 (beta))

**Question:**

A manufacturer runs AI training projects for many business units in a shared AWS account. Operations staff must be able to stop Amazon SageMaker AI training jobs only for their assigned project. Projects and jobs change frequently, so the company wants authorization to adapt without maintaining a separate list of job identifiers in each operator's policy. The central security team maintains authoritative project assignments. Which access-control design BEST meets these requirements?

**Correct answer:** Allow operators to stop training jobs when the job's Project tag matches their centrally assigned principal tag, and prevent operators from changing either authorization tag.

**Explanation:** Tag-based IAM conditions can enforce project-specific access without listing individual jobs. Centrally controlling the principal and resource tags prevents operators from changing the attributes to obtain access to another project's jobs.

### Question 2 (MLA-C02 (beta))

**Question:**

A monitoring application uses an IAM role to check the status of Amazon SageMaker AI endpoints and send test inference requests to one endpoint named fraud-canary. The role's identity-based policy allows sagemaker:DescribeEndpoint on all endpoints and sagemaker:InvokeEndpoint only on the fraud-canary endpoint ARN. A permissions boundary attached to the role allows only sagemaker:DescribeEndpoint on all endpoints.

Status checks succeed, but test inference requests receive AccessDenied errors. No service control policies, resource policies, session policies, or explicit denies affect these requests. The application must retain its status-checking permissions, and the permissions boundary must continue to limit inference access to fraud-canary even if broader identity-based policies are attached later.

Which change meets these requirements?

**Correct answer:** Update the permissions boundary to retain its DescribeEndpoint permission and add sagemaker:InvokeEndpoint only on the fraud-canary endpoint ARN.

**Explanation:** The role's effective permissions are constrained by both its identity-based policies and its permissions boundary. Adding the scoped invocation permission to the boundary enables the existing identity-based allow, preserves status checks, and prevents future identity-based policies from expanding inference access beyond fraud-canary.

### Question 3 (SAP-C03)

**Question:**

A company stores compliance logs in an Amazon S3 bucket in a security account. An existing bucket policy allows the CloudTrail service principal to deliver logs, with permissions restricted to the company's trail by an aws:SourceArn condition. Analyst roles in the security account have identity-based policies that allow them to read the logs.

A new requirement states that all requests to this bucket and its objects made directly with IAM principal credentials must pass through a designated S3 VPC endpoint. The restriction must remain effective even if an administrator later grants additional S3 permissions to an IAM principal. CloudTrail must continue delivering logs directly to the bucket. A separate governance control prevents unauthorized changes to the bucket policy.

Which bucket policy change meets these requirements?

**Correct answer:** Retain the existing CloudTrail allow statement. Add an explicit deny for s3:* on the bucket and its objects, with Principal set to *, that applies when aws:SourceVpce does not equal the designated endpoint ID and aws:PrincipalIsAWSService is false.

**Explanation:** The two conditions are evaluated together. Direct IAM requests outside the designated endpoint are explicitly denied, regardless of additional identity-based allows. Direct requests from the CloudTrail service principal are excluded from this deny by aws:PrincipalIsAWSService. The existing trail-scoped allow continues to authorize CloudTrail delivery.

### Question 4 (AIF-C01)

**Question:**

A company uses AWS Lake Formation to manage access to datasets used for AI development. Data scientists must be able to query approved training tables, but only data stewards may authorize additional users to access those tables. The data scientists currently have no Lake Formation permissions, and the query infrastructure is already configured. Which permission assignment meets these requirements?

**Correct answer:** Give data scientists SELECT permission on the approved tables without the grant option, and reserve access-management permissions for data stewards.

**Explanation:** SELECT allows data scientists to read the approved training data. Withholding the grant option and access-management permissions prevents them from authorizing additional users, preserving the separation between data use and access administration.

### Question 5 (DEA-C01)

**Question:**

A company uses Amazon SageMaker Catalog subscriptions to manage analytics projects' access to Amazon S3 tables. A data owner revokes a project's subscription to a transactions table, and the catalog reports that the subscription's access grant was successfully removed. However, the project's execution role can still query the table through Amazon Athena.

An audit finds that, before the catalog was adopted, the execution role received a separate SELECT grant on all tables in the database through a table wildcard. The role has no other data permissions and cannot read the S3 objects directly. The project must retain its membership, query capabilities, and access to several other approved tables in the database.

Which solution will remove access to the transactions table while preserving the project's approved access?

**Correct answer:** Use AWS Lake Formation to grant the execution role SELECT explicitly on the other approved tables, and then revoke its table-wildcard SELECT grant.

**Explanation:** The independent wildcard grant continues to authorize access after the subscription-specific grant is removed. Granting access to the other approved tables before revoking the wildcard preserves their availability while removing the remaining permission path to the transactions table.

### Question 6 (SCS-C03)

**Question:**

A company uses Amazon CloudFront standard logging (legacy) to deliver security access logs to an Amazon S3 bucket. Logs stop arriving after a hardening deployment replaces the bucket ACL. Viewer requests continue to succeed.

A security engineer verifies the following:
• The distribution still has logging enabled with the correct destination bucket.
• The bucket uses S3 Object Ownership with Bucket owner preferred, so ACLs remain enabled.
• The bucket ACL now grants FULL_CONTROL only to the bucket owner's canonical user.
• The bucket uses SSE-S3 encryption, is in a supported AWS Region, and has no applicable explicit denies.
• S3 Block Public Access is enabled and must remain enabled.

Which change will restore log delivery without changing the logging implementation?

**Correct answer:** Restore a FULL_CONTROL grant for the awslogsdelivery account's canonical user on the destination bucket ACL, while retaining S3 Block Public Access.

**Explanation:** CloudFront standard logging (legacy) requires the awslogsdelivery canonical account to have FULL_CONTROL on the destination bucket ACL. Replacing the ACL removed that required grant. Restoring this specific canonical-user grant enables delivery without making the bucket public or disabling Block Public Access.

### Question 7 (CLF-C02)

**Question:**

A company stores internal documents in Amazon S3 buckets. The security team needs to automatically identify bucket policies that permit access from outside the company's AWS account, even if no external requester has used that access. Which IAM capability should the team use?

**Correct answer:** An IAM Access Analyzer external access analyzer with the account as its zone of trust

**Explanation:** IAM Access Analyzer analyzes supported resource-based policies, including S3 bucket policies, to identify public or cross-account access outside the configured zone of trust. Its findings do not depend on whether that access has been used.

### Question 8 (DVA-C02)

**Question:**

A company uses an Amazon API Gateway HTTP API with an AWS Lambda authorizer. The authorizer uses payload format version 2.0 with simple responses enabled. Its only identity source is $request.header.Authorization, and authorization results are cached for 5 minutes.

The API has GET /summaries and GET /exports routes. Some users can access both routes; others can access only GET /summaries. The authorizer correctly evaluates each user's permissions for the requested route and returns an isAuthorized Boolean. However, a user who successfully requests GET /summaries can then access GET /exports without another authorizer invocation.

Permissions remain unchanged during the cache period. The developer must enforce route-specific permissions while retaining authorizer caching and allowing users to access every route they are permitted to use. The backend integrations must not perform additional authorization checks.

Which change should the developer make?

**Correct answer:** Add $context.routeKey as an additional authorizer identity source, keeping the Authorization header identity source and simple responses enabled.

**Explanation:** API Gateway constructs the authorizer cache key from the configured identity sources. Including the route key separates cached decisions for different routes, so an Allow for GET /summaries is not reused for GET /exports. Requests with the same token to the same route can still reuse cached results.

### Question 9 (DOP-C02)

**Question:**

An energy company publishes machine-readable operating limits in Amazon S3. Customers must be able to retrieve the files anonymously. The files contain no personal or confidential information, but an unauthorized change could cause customers' equipment to operate unsafely.

The company's deployment automation currently assigns every publicly readable dataset a low-risk handling profile. This profile permits application administrators to overwrite objects and records only bucket-level configuration changes. A security review requires operating-limit files to remain publicly readable while preventing writes outside a data-owner-approved release pipeline. Every successful object write and deletion must have a retained audit record, and previous file versions must be retained for recovery.

Which change should a DevOps engineer implement?

**Correct answer:** Record owner-approved public-confidentiality and critical-integrity attributes in the authoritative dataset inventory. Map that profile to automated controls that restrict object mutations to the approved release role, enable S3 Versioning, and retain CloudTrail S3 write data events in a protected audit destination.

**Explanation:** The classification must account for the consequences of modification independently of disclosure. This profile preserves anonymous reads while enforcing pipeline-only mutations, retaining earlier versions, and auditing object-level writes and deletions. CloudTrail S3 write data events capture object mutations; bucket-level management events alone do not.

### Question 10 (SAA-C03)

**Question:**

A company operates an application in two AWS Regions. Each deployment must decrypt records locally without depending on the availability of the other Region. The records are encrypted with a symmetric multi-Region AWS KMS key that has a primary key in one Region and a replica key in the other Region.

Both deployments use the same IAM role. The role's IAM policy allows kms:Decrypt on both key ARNs. Decryption succeeds with the primary key but fails with an access denied error when the application explicitly specifies the replica key ARN. The primary key policy enables the account to delegate key-use permissions through IAM policies. The replica key policy allows only a security administrator role to administer the key and does not enable IAM delegation. Both keys are enabled, and no KMS grants or explicit denies apply.

Which change will allow both deployments to meet the requirements with least privilege?

**Correct answer:** Add a statement to the replica key policy that allows the application role to perform kms:Decrypt on that key.

**Explanation:** Related multi-Region KMS keys share key material but have independently managed key policies. Explicitly authorizing the application role to decrypt with the replica resolves the missing permission while preserving local operation and limiting access to the required action and key.

### Question 11 (ANS-C01)

**Question:**

A company uses a corporate SAML identity provider to federate network engineers into a NetworkOperator role in a central tooling account. An existing command-line tool runs with this role's temporary credentials and calls AWS STS AssumeRole to obtain credentials for a RouteMaintenance role in each workload account. The tool then creates, replaces, or deletes routes in approved VPC route tables across two AWS Regions. Route table IDs are supplied to the tool.

The workload-account roles have not yet been configured. The NetworkOperator role currently has only read-only Amazon EC2 permissions. The company must preserve the existing federation and tool workflow, use no long-lived credentials, and grant only the required route-management access. No service control policies, permissions boundaries, or session policies restrict the required actions.

Which configuration meets these requirements?

**Correct answer:** Allow NetworkOperator to call sts:AssumeRole on the designated RouteMaintenance role ARNs. Configure each RouteMaintenance role's trust policy to trust NetworkOperator for sts:AssumeRole. Attach permissions to RouteMaintenance allowing ec2:CreateRoute, ec2:ReplaceRoute, and ec2:DeleteRoute on the approved route table ARNs, restricted to the required Regions.

**Explanation:** Cross-account role assumption requires authorization for the caller and an appropriate trust policy on the destination role. After assuming RouteMaintenance, the tool uses that role's permissions—not NetworkOperator's permissions—to change routes. Scoping the route actions to approved route table ARNs and the required Regions provides least-privilege access while preserving temporary credentials and the existing workflow.

## Practice Access Control in a full exam

Questions on Access Control in each practice exam, with every answer option and an explanation for each:

- [SAA-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate): 22 questions
- [DOP-C02](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional): 10 questions
- [AIB-C01 (beta)](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist): 9 questions
- [DEA-C01](https://blowstack.com/practice-exams/aws-certified-data-engineer-associate): 8 questions
- [SAP-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional): 6 questions
- [SCS-C03](https://blowstack.com/practice-exams/aws-certified-security-specialty): 6 questions
- [DVA-C02](https://blowstack.com/practice-exams/aws-certified-developer-associate): 6 questions
- [CLF-C02](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner): 5 questions
- [AIF-C01](https://blowstack.com/practice-exams/aws-certified-ai-practitioner): 4 questions
- [MLA-C02 (beta)](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate): 2 questions
- [ANS-C01](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty): 2 questions

## What is Access Control

Access control answers a practical question: may this requester perform this action on this resource in this context? A requester can be a person, application, workload role, or cloud service. Authentication establishes identity; authorization evaluates whether a requested operation is permitted. Successful sign-in does not imply permission to read a database, change a firewall, or view security logs.

Cloud platforms commonly express permissions through policies attached to identities or resources. Roles collect permissions that users or workloads can assume, while attributes such as resource tags, organization membership, source network, or session characteristics can influence decisions. For example, a deployment role might update an application but lack permission to read its production secrets. Temporary role credentials reduce dependence on long-lived access keys, although their permissions and session duration still require careful design.

Least privilege means granting only the actions, resources, and conditions required for a task. In AWS, requests are implicitly denied unless applicable policies allow them, and an applicable explicit deny overrides an allow. Effective authorization can also depend on permissions boundaries, session policies, organization policies, and resource policies. A permissions boundary or service control policy limits possible permissions; it does not grant permissions by itself. Cross-account access often requires both permission for the caller and an appropriate trust or resource policy in the destination account.

Infrastructure location and authorization are separate concerns. An AWS Region is a geographic area containing multiple Availability Zones, which provide separate infrastructure locations for resilient deployments. Edge locations serve functions such as content delivery closer to viewers and are not Availability Zones. IAM is a global service within an AWS partition, while many resources and their permissions reference a particular Region. Deploying across Availability Zones improves availability but does not automatically change access rights. Likewise, selecting a Region does not create a geographic access restriction; enforce such requirements with appropriate policies and service-specific controls.

Access control operates at multiple layers. Network rules determine whether traffic can reach an endpoint, while application and identity policies determine what requests are accepted. A reachable API can still reject an unauthorized request, and a correctly authorized request can fail because routing or firewall rules block connectivity. Public content delivered through an edge location also does not require a publicly accessible origin: a distribution can authorize origin requests while restricting viewers through signed URLs or cookies.

Monitoring pipelines need their own permissions. A Lambda function can execute successfully yet fail to write logs if its execution role lacks the necessary CloudWatch Logs permissions. API Gateway logging requires the appropriate logging configuration and delivery permissions for the API type. CloudFront log delivery requires compatible destination settings and permissions for the selected logging mechanism. Check encryption-key permissions where relevant, destination policies, account and Region selection, and whether the relevant logging feature is enabled before interpreting missing records as an absence of activity.

Health checks illustrate the difference between connectivity and authorization. A health-check request can fail because a firewall blocks the checker, TLS settings are incompatible, or the endpoint requires credentials the checker does not supply. Prefer a narrowly scoped health endpoint that reveals little information rather than weakening authorization for the entire application. Similarly, give log producers permission to write only to intended destinations, and separate log-reading privileges from permissions to delete logs or disable collection.

The main trade-off is precision versus operational complexity. Broad permissions simplify initial setup but increase the damage possible from mistakes or compromised credentials. Highly restrictive policies can break deployments and obscure monitoring failures. Use tested policy changes, access-denied evidence, policy analysis, and periodic permission reviews to refine access. Audit logs provide evidence of activity, but they are not themselves an enforcement mechanism, and an administrator label does not guarantee unrestricted access through every policy layer.

## Learn Access Control in lessons

14 short lessons on Access Control in 9 certification courses.

- AIB-C01 (beta): Access Control: scoped permissions (Premium), https://app.blowstack.com/lessons/access-control-scoped-permissions-aib-c01
- MLA-C02 (beta): Access Control: least-privilege roles for ML systems (Premium), https://app.blowstack.com/lessons/access-control-least-privilege-roles-for-ml-systems-mla-c02
- SAP-C03: Access Control: authorization across delegation boundaries (Premium), https://app.blowstack.com/lessons/access-control-authorization-across-delegation-boundaries-sap-c03
- AIF-C01: Access Control: least-privilege access to data (Premium), https://app.blowstack.com/lessons/access-control-least-privilege-access-to-data-aif-c01
- DEA-C01: Access Control: granting access to shared data (Premium), https://app.blowstack.com/lessons/access-control-granting-access-to-shared-data-dea-c01
- SCS-C03: Access Control: service-to-service authorization (Premium), https://app.blowstack.com/lessons/access-control-service-to-service-authorization-scs-c03
