# Least privilege

Least privilege means granting each identity or workload only the access required for its approved tasks, limited to the necessary resources, conditions, and duration.

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

> Least privilege reduces the damage that compromised credentials, software defects, and operational mistakes can cause. In cloud environments, it applies to identity permissions, workload roles, administrative access, network connectivity, and organizational guardrails, and requires ongoing review as requirements change.

Practice exams covering Least privilege: [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), [SOA-C03 AWS Certified CloudOps Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate), [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), [AIP-C01 AWS Certified Generative AI Developer - Professional](https://blowstack.com/practice-exams/aws-certified-generative-ai-developer-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).

## 7 key aspects of Least privilege

Least privilege combines narrowly scoped permissions, limited access duration, layered controls, escalation safeguards, and continuous validation.

1. **Task-scoped permissions**: Grant only the actions and resources required for a defined responsibility. Separate deployment, runtime, investigation, and recovery roles so that one compromise does not expose every operational capability.
2. **Limited duration and controlled delegation**: Prefer temporary workload and user credentials, with session duration matched to the task. Treat presigned URLs and other bearer capabilities as sensitive delegated access, and restrict both their scope and lifetime.
3. **Effective-access evaluation**: Review identity policies, resource policies, trust relationships, session restrictions, and organizational controls together. Account for explicit denies and principal-specific resource-policy behavior rather than judging access from a single attached policy.
4. **Layered access restrictions**: Constrain identity permissions, network reachability, application authorization, and operating-system privileges independently. A network connection, CORS rule, or WAF rule does not establish that an operation is authorized.
5. **Guardrails against privilege escalation**: Use permissions boundaries and organization policies to constrain delegated administration. Restrict capabilities such as passing roles, editing policies, and modifying trust relationships because these can enable indirect escalation.
6. **Prepared privileged access**: Provision and test restricted incident-response and break-glass access before it is needed. Protect root and administrative access with strong authentication, auditability, and controlled elevation instead of routine unrestricted use.
7. **Continuous review and refinement**: Compare granted permissions with intended responsibilities, observed activity, and access-analysis findings. Remove unused or unintended access while testing infrequent but necessary operations to avoid breaking recovery procedures.

## Cloud services for Least privilege

IAM defines narrowly scoped roles and policies, and STS turns those roles into temporary sessions rather than permanent credentials. AWS Organizations adds central guardrails that constrain supported access even when member-account administrators manage local permissions. IAM Access Analyzer helps teams discover unintended sharing, unused access, and policy issues so they can refine those permissions. Systems Manager provides controlled administrative access to managed compute, while Verified Access narrows access to supported applications using identity and device context. Together, these services support permission scope, access duration, administrative control, and ongoing review; none replaces application authorization, appropriate network restrictions, or tested incident-response procedures.

- **[AWS Organizations](https://blowstack.com/tools/cloud-services-hub/aws-organizations)**: AWS Organizations provides centrally managed permission guardrails through service control policies and resource control policies. These policies constrain supported access without granting permissions and must be designed with their documented scope and exceptions in mind.
- **[AWS Systems Manager](https://blowstack.com/tools/cloud-services-hub/aws-systems-manager)**: Systems Manager Session Manager supports controlled administrative sessions without requiring inbound SSH or RDP access. Least privilege depends on restricting session initiators, target nodes, session documents, and the permissions assigned to managed nodes.

## Least privilege exam questions

Real practice exam questions on Least privilege, with the correct answer and its explanation.

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

**Question:**

An insurer runs an AI claims-processing application on Amazon EC2. Operations staff normally have no permission to change production resources. During an approved incident, one engineer must view health metrics and reboot two designated inference instances for up to 90 minutes. The engineer must not change instance configurations, access claims datasets, or retain production-change access after the incident window. Which access arrangement should the governance lead approve?

**Correct answer:** Grant an approval-controlled IAM role session lasting no more than 90 minutes, with only the required metric-viewing permissions and reboot permission for the two designated instances. Require a new approval for any subsequent elevated session.

**Explanation:** This arrangement limits the permitted actions, target resources, and access duration to the approved incident requirements. Requiring fresh approval prevents the engineer from simply obtaining another elevated session after the window ends.

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

**Question:**

A company shares proprietary model artifacts with external evaluators through Amazon S3 presigned download URLs. A signing application uses an IAM role that allows s3:GetObject only on objects under s3://evaluation-artifacts/approved/. The application currently generates URLs with a 1-hour expiration, and its configuration cannot be changed during the current release.

A security requirement states that each URL must work for the first 10 minutes after signing and must be rejected afterward, regardless of its configured expiration. The restriction must apply automatically to all presigned downloads under approved/ without affecting requests that use an Authorization header. The signing credentials remain valid for longer than 1 hour, and no other policies restrict these downloads.

Which bucket policy change meets these requirements?

**Correct answer:** Add a Deny statement for Principal * and s3:GetObject on arn:aws:s3:::evaluation-artifacts/approved/*, with StringEquals requiring s3:authType to be REST-QUERY-STRING and NumericGreaterThan requiring s3:signatureAge to be 600000.

**Explanation:** S3 measures s3:signatureAge in milliseconds, so 600000 represents 10 minutes. The explicit Deny overrides the signing role's Allow for older presigned requests. The authentication-type condition limits the restriction to query-string authentication, leaving Authorization-header requests unaffected. Younger URLs remain authorized by the existing role policy.

### Question 3 (SAP-C03)

**Question:**

A company uses AWS CloudFormation to deploy an EC2-based payment-processing application and its instance role. An existing AWS Systems Manager Parameter Store hierarchy contains three SecureString parameters:

• /payments/prod/runtime/vendor-token
• /payments/prod/runtime/database-password
• /payments/prod/runtime/admin-password

The application needs the first two parameters. The third parameter is reserved for a separate administrative service. All three parameters must remain at their current paths. The application’s parameter-loading code can be changed to use different Parameter Store retrieval APIs.

A security review requires that compromised application credentials cannot retrieve the administrative parameter, even if an attacker changes the API requests. The instance role currently has no Parameter Store permissions. Decryption permissions are already configured, and no other policies grant the role access to parameters.

Which change should a solutions architect make to the CloudFormation template and application?

**Correct answer:** Allow ssm:GetParameters on only the complete ARNs of the vendor-token and database-password parameters, without granting other Parameter Store read actions. Configure the application to request those two parameters explicitly with decryption enabled.

**Explanation:** GetParameters supports resource-level permissions for individual parameters. Restricting the action to the two required parameter ARNs lets the application retrieve its credentials without authorizing access to the administrative parameter. An attacker cannot expand that access merely by changing the requested names or choosing another retrieval API.

### Question 4 (AIF-C01)

**Question:**

A company runs an AI document-classification application on Amazon EC2. The application needs to read objects only from the approved-input/ prefix in an Amazon S3 bucket that also contains confidential documents outside that prefix. The application must not modify any objects or access other documents. Company policy also prohibits storing long-term AWS access keys on the instance. Which solution best meets these requirements?

**Correct answer:** Attach an IAM role to the EC2 instance through an instance profile. Allow s3:GetObject only for objects under the approved-input/ prefix.

**Explanation:** An EC2 instance profile supplies temporary credentials for the attached IAM role. A policy allowing only s3:GetObject on the specified prefix provides the required read access without granting access to other documents or permission to modify objects.

### Question 5 (SOA-C03)

**Question:**

A company allows CloudOps engineers to launch Amazon EC2 instances by using an OperationsLauncher IAM role. The engineers must attach an existing instance profile named AppInstanceProfile, which contains the AppRuntime IAM role. AppRuntime already trusts the EC2 service. OperationsLauncher has all required EC2 permissions, but instance launches that specify this profile fail because OperationsLauncher is not authorized to pass the role.

The engineers must be able to pass only AppRuntime, and only to Amazon EC2. They must not receive permission to assume AppRuntime directly. No other policies grant role-passing permissions or restrict the required operations.

Which change should a CloudOps engineer make?

**Correct answer:** Add a policy to OperationsLauncher that allows iam:PassRole on the AppRuntime role ARN, with a StringEquals condition requiring iam:PassedToService to equal ec2.amazonaws.com.

**Explanation:** This grants the launching identity permission to pass only AppRuntime. The iam:PassedToService condition limits the receiving service to EC2. It does not grant sts:AssumeRole, so engineers receive no permission to assume AppRuntime directly.

### Question 6 (AIP-C01)

**Question:**

A company runs foundation model evaluation workers that read Amazon S3 datasets by using temporary credentials for an IAM role named EvaluationWorker. The role and bucket are in the same AWS account.

The role's identity policy allows s3:GetObject on both approved/* and restricted/* in the dataset bucket. A permissions boundary allows s3:GetObject only on approved/* and contains no explicit Deny statements. However, a legacy bucket policy directly grants s3:GetObject on restricted/* to the assumed-role session ARN used by the workers. No other policies grant access to these objects.

A security test confirms that a worker can read restricted/* despite the permissions boundary. All worker requests already traverse the company's approved S3 VPC endpoint. The company requires compromised workers to be unable to read restricted/* while retaining access to approved/*.

Which change meets these requirements?

**Correct answer:** Remove the bucket policy statement that grants the worker's assumed-role session access to restricted/*.

**Explanation:** Removing the direct session grant eliminates the bypass. The remaining identity-based permission is limited by the permissions boundary, so approved/* remains readable and restricted/* is implicitly denied. The scenario specifies that no alternative grants exist.

### Question 7 (CLF-C02)

**Question:**

A company runs two AWS Lambda functions. One function only writes order records to a designated Amazon DynamoDB table. The other function only publishes notifications to a designated Amazon SNS topic. The functions currently share an IAM execution role that permits both operations. Which change would best apply the principle of least privilege?

**Correct answer:** Assign separate execution roles, granting each role only its function's required operation on its designated resource.

**Explanation:** Separate, narrowly scoped roles ensure that each function receives only the permissions it needs. The order function can write to the designated table, and the notification function can publish to the designated topic.

### Question 8 (DEA-C01)

**Question:**

A data platform stores encrypted data keys alongside application-encrypted datasets. A migration job must use the AWS KMS ReEncrypt API to move the existing encrypted data keys from a legacy customer managed symmetric KMS key to a replacement customer managed symmetric KMS key in the same account and Region.

The job must not receive plaintext data keys, encrypt new data keys, or re-encrypt data keys in the reverse direction. Both KMS key policies enable the account to delegate permissions through IAM policies. The job's IAM role currently has no KMS permissions, and no other policies or grants authorize its use of either key.

Which custom IAM policy configuration meets these requirements with the least privilege?

**Correct answer:** Allow kms:ReEncryptFrom on the legacy key ARN, and allow kms:ReEncryptTo on the replacement key ARN.

**Explanation:** KMS checks ReEncryptFrom permission on the source key and ReEncryptTo permission on the destination key. These narrowly scoped permissions allow only the required migration direction. KMS performs the re-encryption internally without returning plaintext to the job, and neither permission authorizes standalone Encrypt or Decrypt calls.

### Question 9 (SCS-C03)

**Question:**

A security engineer investigates an unauthorized assumption of a privileged IAM role in a production account. The role trusts a specific IAM user, ReleaseEngineer, in a tooling account. ReleaseEngineer's identity policy allows sts:AssumeRole on the production role.

The production role's only trust policy statement is:

{"Effect":"Allow","Principal":{"AWS":"arn:aws:iam::111122223333:user/ReleaseEngineer"},"Action":"sts:AssumeRole","Condition":{"BoolIfExists":{"aws:MultiFactorAuthPresent":"true"}}}

CloudTrail confirms that ReleaseEngineer successfully assumed the role by using long-term access keys without supplying MFA parameters. No other authorization controls restrict the request.

The company requires every new assumption of this role to use MFA-authenticated credentials or a valid MFA code supplied with the AssumeRole request. Requests with a false or missing MFA context must fail. The trusted user and the role's workload permissions must remain unchanged.

Which change meets these requirements?

**Correct answer:** Replace the trust policy's BoolIfExists condition with {"Bool":{"aws:MultiFactorAuthPresent":"true"}}.

**Explanation:** Bool requires the MFA context key to be present and true for this Allow statement to match. It rejects both missing and false MFA context while preserving assumptions made with valid MFA authentication. BoolIfExists allowed the original request because the key was absent.

### Question 10 (DVA-C02)

**Question:**

A developer uses an IAM deployment role to update an existing AWS Lambda function named invoice-processor. The developer must configure the function to use an existing execution role named InvoiceRuntimeRole in the same AWS account. The deployment role already allows lambda:UpdateFunctionConfiguration on the function but has no other permissions. InvoiceRuntimeRole already trusts the Lambda service and has the required runtime permissions. A security requirement states that the deployment role must be able to assign only InvoiceRuntimeRole, and only to Lambda. Which change meets these requirements with the least privilege?

**Correct answer:** Attach a policy to the deployment role allowing iam:PassRole on arn:aws:iam::123456789012:role/InvoiceRuntimeRole, with a StringEquals condition requiring iam:PassedToService to be lambda.amazonaws.com.

**Explanation:** iam:PassRole authorizes the deployment role to assign an execution role to an AWS service. The specific role ARN restricts which role can be passed, and iam:PassedToService restricts the destination service to Lambda.

### Question 11 (DOP-C02)

**Question:**

A company uses an automation IAM role in each of 100 AWS accounts to restart approved application services on EC2 instances through AWS Systems Manager Run Command. Each account contains a centrally maintained, untagged custom SSM document named RestartApprovedService. The document runs a fixed restart procedure and does not accept arbitrary commands.

The automation must be allowed to call ssm:SendCommand only with this document and only against instances tagged Environment=Production. Requests targeting untagged instances or instances with any other Environment value must fail. The automation cannot modify documents or instance tags. No other policies grant ssm:SendCommand, and no other authorization controls block the required requests.

A DevOps engineer will deploy an IAM policy template into every account, substituting the local account ID and Region. Which policy design meets these requirements?

**Correct answer:** Create two Allow statements for ssm:SendCommand. Scope one statement to the exact RestartApprovedService document ARN without a tag condition. Scope the other statement to the account's EC2 instance ARN pattern with StringEquals for ssm:resourceTag/Environment set to Production.

**Explanation:** SendCommand requires authorization for the document and the targeted instances. Separate statements authorize only the approved document while applying the Environment condition specifically to instances. An untagged or nonproduction instance receives no applicable Allow, and no other document is authorized.

### Question 12 (SAA-C03)

**Question:**

A company runs application instances and PostgreSQL database instances in separate subnets of an Amazon VPC. Only application instances occupy the application subnet, 10.40.10.0/24. Route tables send all traffic between the two tiers through a security appliance in a third subnet. The appliance preserves the original source IP addresses and permits TCP port 5432.

The application instances' security group allows outbound TCP port 5432 to the database subnet. The database instances' security group allows inbound TCP port 5432 with the application instances' security group as the source. Network ACLs allow all required traffic, and routing through the appliance has been verified. However, new database connections from the application instances time out.

The company must retain appliance inspection and restrict database access to the application subnet. Which change will meet these requirements?

**Correct answer:** Replace the database security group's inbound rule with an allowance for TCP port 5432 from 10.40.10.0/24.

**Explanation:** When traffic is routed through a middlebox, referencing the originating instances' security group does not authorize that forwarded traffic. The database rule must instead identify the originating private IP addresses or their CIDR range. Allowing the dedicated application subnet permits the required connections without bypassing inspection or allowing other subnets.

### Question 13 (ANS-C01)

**Question:**

A company has application servers and API servers in VPCs owned by different AWS accounts in the same AWS Region. The VPCs have an active peering connection and correct routes. Application servers initiate HTTPS connections to the API servers by using private IP addresses and TCP source ports 49152–65535.

The API servers' security group allows inbound TCP port 443 from the application servers' security group through a cross-account security group reference. The API security group has no outbound rules. The application security group permits the outbound connections.

The application subnet's network ACL permits the requests and responses. The API subnet's network ACL permits inbound TCP port 443 from the application subnet. Its outbound rules contain rule 100, which denies all traffic to the application VPC CIDR, followed by the default deny rule. Packet captures show that requests reach the API servers and that the API servers generate responses, but the application servers do not receive the responses.

A security requirement prohibits API servers from initiating connections to the application servers. Which change will restore HTTPS connectivity while preserving this requirement?

**Correct answer:** Add outbound rule 90 to the API subnet's network ACL to allow TCP destination ports 49152–65535 to the application subnet CIDR. Leave the API security group's outbound rules empty.

**Explanation:** Network ACLs are stateless, so the API subnet requires an outbound rule for responses destined for the application servers' ephemeral ports. Rule 90 is evaluated before deny rule 100. Security groups are stateful and permit responses to allowed inbound connections without outbound rules. Keeping the API security group's outbound rules empty prevents API-initiated connections, even though the network ACL permits the response port range.

## Practice Least privilege in a full exam

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

- [SCS-C03](https://blowstack.com/practice-exams/aws-certified-security-specialty): 72 questions
- [CLF-C02](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner): 32 questions
- [SAA-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate): 23 questions
- [DOP-C02](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional): 16 questions
- [AIP-C01](https://blowstack.com/practice-exams/aws-certified-generative-ai-developer-professional): 15 questions
- [DEA-C01](https://blowstack.com/practice-exams/aws-certified-data-engineer-associate): 10 questions
- [AIB-C01 (beta)](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist): 9 questions
- [SAP-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional): 9 questions
- [MLA-C02 (beta)](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate): 4 questions
- [SOA-C03](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate): 3 questions
- [ANS-C01](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty): 3 questions
- [AIF-C01](https://blowstack.com/practice-exams/aws-certified-ai-practitioner): 2 questions
- [DVA-C02](https://blowstack.com/practice-exams/aws-certified-developer-associate): 1 question

## What is Least privilege

Least privilege is an authorization design principle: an identity should be able to perform its intended work without receiving unrelated capabilities. Effective access is constrained by actions, resources, conditions, and time. For example, an application that reads invoices should receive permission to read objects under the required storage prefix, not permission to delete objects, modify bucket policies, or access every bucket. Authentication establishes who is requesting access; authorization determines whether that identity may perform a particular operation.

Start with documented access requirements and separate responsibilities into distinct roles. Scope permissions to supported resource identifiers and use conditions where they meaningfully restrict access, such as approved organization membership, resource tags, or a required network endpoint. Some API actions do not support resource-level permissions and therefore require a wildcard resource. In those cases, restrict the action set and use supported conditions rather than assuming that every wildcard is avoidable. Test both required operations and operations that must be denied.

Cloud permissions must be evaluated across all applicable policy types, not just the policy attached to a user or role. In AWS, identity policies, resource policies, role trust policies, permissions boundaries, session policies, and organization policies can influence the result. Explicit denies in applicable policies override allows. Permissions boundaries and session policies limit permissions rather than independently granting them. However, resource-based grants can behave differently depending on whether they target a user, role, or role session, so policy evaluation is not always a simple intersection of every policy. Investigating unintended access requires examining the actual principal, request context, and grant path.

Prefer temporary credentials over embedded long-lived access keys. AWS STS issues credentials for an assumed role, with permissions constrained by the role and any applicable session policy. A short session lifetime reduces the exposure window but does not make an overprivileged role safe. Amazon S3 presigned URLs delegate a particular signed request, such as downloading a specific object, and should use an appropriately limited signer and expiration. They are bearer capabilities: anyone possessing a valid URL can use it. A URL signed with temporary credentials cannot remain valid beyond those credentials, and applicable policy changes can also prevent access.

Compute workloads should obtain credentials through dedicated workload identities. An EC2 instance profile supplies a role to an instance; a Lambda execution role authorizes function code; a service role allows an AWS service to act on a customer's behalf. These mechanisms are not interchangeable, and each needs an appropriately restricted trust policy and permissions policy. For example, deployment permissions should be separate from runtime permissions. Restrict iam:PassRole to approved roles and intended services so that a developer cannot attach a more powerful role to a workload and indirectly escalate privileges.

Administrative access needs both identity and network controls. AWS Systems Manager Session Manager can provide role-controlled access without opening inbound SSH or RDP ports, but requires suitable instance permissions and connectivity to the service. Limit who can start sessions, which managed nodes they can target, and which session documents they can use. Configure logging where supported; SSH and port-forwarding sessions do not provide the same command visibility as standard shell sessions. EC2 Instance Connect provides temporary SSH-key delivery, but does not eliminate the need to restrict SSH reachability, operating-system users, and host-level privileges.

Apply least privilege to network paths as well as API calls. Permit only the necessary source, destination, protocol, and port combinations, and review unnecessary ingress and egress. VPC Network Access Analyzer can identify network paths that violate defined access requirements. AWS Verified Access can enforce identity- and device-context policies for supported applications without broadly exposing the surrounding network. Evaluate network exposure alongside Amazon Inspector workload findings, while recognizing that reachability, vulnerability assessment, and application authorization answer different questions.

At the network edge, restrict direct origin access and validate requests at the appropriate layer. CloudFront origin access control can authorize access to a private S3 origin; configured headers and origin checks can support other origin-protection designs. AWS WAF rules help filter malicious requests, including patterns associated with OWASP Top 10 threats, but are not a substitute for application authorization. S3 CORS controls browser cross-origin behavior and does not grant object permissions or protect an otherwise public bucket. AWS IoT policies should limit devices to required operations and topic resources. Shield Advanced provides DDoS protections and response capabilities, not identity authorization.

Use organization-level controls to prevent local administrators from exceeding centrally defined limits. AWS Organizations service control policies constrain supported permissions for principals in member accounts, while resource control policies constrain access to supported resources. Neither policy type independently grants access. Understand their coverage and exceptions rather than treating them as universal controls. AI services opt-out policies address the use of content for AI service improvement, and declarative policies govern supported service configurations; these are distinct from IAM permission grants.

Prepare incident access before an incident occurs. Provision and test narrowly scoped response roles for evidence collection, containment, and recovery, separating disruptive actions from routine investigation. Avoid permanent administrator access merely to make response easier. Time-bounded elevation, protected approval paths, and audited break-glass procedures preserve response speed while limiting blast radius. Protect root access with appropriate MFA and tightly controlled recovery processes. Where centralized root access is enabled for AWS member accounts, remove unnecessary member-account root credentials and use supported, task-scoped privileged root sessions instead of retaining standing root access.

Least privilege is an iterative process, not a one-time policy-writing exercise. IAM Access Analyzer can identify external access, analyze supported internal access, detect unused access, validate policies, and generate policy suggestions from supported activity. An external-access finding describes a policy-enabled access path, not proof that someone used it. Review findings against intended sharing and remediate unnecessary grants. Activity-derived policies are useful starting points, but observation windows may miss rare maintenance and recovery operations. Combine analysis, access logs, owner review, and regression tests. The trade-off is additional policy maintenance and occasional access failures; manage this through reusable role patterns and controlled escalation rather than broad permanent permissions.

## Learn Least privilege in lessons

15 short lessons on Least privilege in 11 certification courses.

- AIB-C01 (beta): Least privilege: scoping AI system permissions (Premium), https://app.blowstack.com/lessons/least-privilege-scoping-ai-system-permissions-aib-c01
- MLA-C02 (beta): Least privilege: scoping ML artifact access (Premium), https://app.blowstack.com/lessons/least-privilege-scoping-ml-artifact-access-mla-c02
- SAP-C03: Least privilege: restricting cross-account role assumption (Premium), https://app.blowstack.com/lessons/least-privilege-restricting-cross-account-role-assumption-sap-c03
- AIF-C01: Least privilege: scoped permissions for AI workloads (free), https://app.blowstack.com/lessons/least-privilege-scoped-permissions-for-ai-workloads-aif-c01
- SOA-C03: Least privilege: scoping IAM access (free), https://app.blowstack.com/lessons/least-privilege-scoping-iam-access-soa-c03
- AIP-C01: Least privilege: permission scope for AI integrations (Premium), https://app.blowstack.com/lessons/least-privilege-permission-scope-for-ai-integrations-aip-c01
