Study guide
9.1 Complete AWS Policy Evaluation Logic & Decision Flow
AWS Certified Security – Specialty SCS-C03
Track your progress as you read
Finished sections get a check mark.
1 Chapter 0: Getting Started with the SCS-C03 Exam
0.1 SCS-C03 Exam Format, Domains & Study Strategy
2 Chapter 1: Security Monitoring, Alerting & Threat Detection Services
1.1 Workload Monitoring & Health Check Strategies 1.2 Amazon GuardDuty Architecture & Finding Types 1.3 AWS Security Hub & Automated Conformance Assessment 1.4 Amazon Macie & Automated Data Classification
3 Chapter 2: Enterprise Logging Architecture & Analysis
2.1 Multi-Account CloudTrail & Organization Trails 2.2 Amazon Security Lake & OCSF Log Normalization 2.3 Network Telemetry: VPC Flow Logs & Route 53 Resolver Logs 2.4 Troubleshooting Security Telemetry & Log Ingestion 2.5 Analyzing, Normalizing & Correlating Security Logs
4 Chapter 3: Incident Readiness, Response Planning & Automated Runbooks
3.1 Incident Response Planning & Blast Radius Mitigation 3.2 Testing & Validating Incident Response Plans 3.3 Automated Incident Remediation with EventBridge & Step Functions 3.4 Systems Manager OpsCenter & Incident Manager Runbooks
5 Chapter 4: Digital Forensics, Evidence Preservation & Root Cause Analysis
4.1 Forensic Artifact Capture & Chain of Custody 4.2 Threat Containment & Network Quarantine Techniques 4.3 Correlating Events & Root Cause Analysis with Amazon Detective
6 Chapter 5: Edge Protection, WAF & DDoS Mitigation
5.1 CloudFront Security Headers & Origin Access Control 5.2 AWS WAF Architecture & Advanced Rule Sets 5.3 AWS Shield Advanced & DDoS Defense Architectures 5.4 Edge Integrations: Managed Rule Groups, OCSF Ingestion, IoT Policies & CORS
7 Chapter 6: Compute Workload Hardening, Vulnerability Management & GenAI Security
6.1 Golden AMIs & Container Image Hardening Pipelines 6.2 Amazon Inspector Vulnerability Management & Patch Automation 6.3 Generative AI Security: Amazon Bedrock Guardrails & Model Protection 6.4 Instance Profiles, Service Roles & Execution Roles for Compute 6.5 Secure Administrative Access: Session Manager & EC2 Instance Connect
8 Chapter 7: Network Security, Segmentation, Private Connectivity & Encryption in Transit
7.1 Layered Traffic Filtering: SGs, NACLs & AWS Network Firewall 7.2 Hybrid Connectivity Security: VPN, Direct Connect & MACsec 7.3 Zero-Trust Access & Network Access Analysis 7.4 Enforcing Encryption in Transit: ELB Security Policies, TLS & HTTPS-Only Access 7.5 Private Access: VPC Endpoints, PrivateLink, Client VPN & Verified Access 7.6 Inter-Resource Encryption in Transit: Nitro, EMR, EKS & SageMaker AI
9 Chapter 8: Enterprise Authentication, Federation & Directory Services
8.1 AWS IAM Identity Center & Enterprise IdP Integration 8.2 Amazon Cognito User Pools & Identity Pools 8.3 AWS Security Token Service (STS) & Temporary Credential Workflows 8.4 Troubleshooting Identity & Authentication Failures
10 Chapter 9: Advanced Authorization Architecture & Policy Evaluation
9.1 Complete AWS Policy Evaluation Logic & Decision Flow 9.2 Permission Boundaries & Safe Admin Delegation 9.3 Attribute-Based Access Control (ABAC) with IAM Tags
11 Chapter 10: Least Privilege, Policy Validation & External Access Auditing
10.1 IAM Roles Anywhere & Amazon Verified Permissions 10.2 Auditing External & Unused Access with IAM Access Analyzer 10.3 Analyzing Authorization Failures & Policy Simulation
12 Chapter 11: Cryptographic Services: KMS Architecture & Key Governance
11.1 AWS KMS Architecture, Key Policies & Key Types 11.2 KMS Grants, Encryption Context & Multi-Region Keys 11.3 Imported Key Material (BYOK) & CloudHSM Key Stores
13 Chapter 12: Secrets Management, Sensitive Data Masking & PKI
12.1 AWS Secrets Manager & Systems Manager Parameter Store 12.2 Sensitive Data Masking in Telemetry & Messaging 12.3 AWS Certificate Manager (ACM) & Private CA Architecture
14 Chapter 13: Storage Encryption, Integrity, Lifecycle & Backup Protection
13.1 S3 Encryption Modes & S3 Bucket Keys 13.2 Data Integrity Protection: Object Lock, Vault Lock & Code Signing 13.3 Secure Data Lifecycle, Cross-Account Replication & Backup Vault Lock 13.4 Lifecycle, Backup & Secure Replication Beyond S3: EFS, FSx, DLM, DataSync & Air-Gapped Vaults
15 Chapter 14: Multi-Account Architecture, Organizations & Control Tower Governance
14.1 AWS Organizations Structure, Hierarchy & Resource Control 14.2 AWS Control Tower Landing Zone & Guardrail Framework 14.3 Delegated Administrator Architecture & Root User Governance 14.4 Centralized Resource Sharing with RAM & Service Catalog
16 Chapter 15: Compliance Automation, Infrastructure as Code Security & Auditing
15.1 Infrastructure as Code Security & Multi-Account Deployment 15.2 AWS Config, Conformance Packs & Automated Remediation 15.3 Audit Evidence Collection & Well-Architected Security Reviews
9.1 Complete AWS Policy Evaluation Logic & Decision Flow
Key Takeaways Explain This Topic AI
AWS authorization evaluation follows a strict deterministic algorithm starting with an implicit default deny that requires an explicit allow in an applicable policy.
An explicit Deny statement in any applicable policy (SCP, RCP, IAM identity policy, resource-based policy, permission boundary, or session policy) immediately terminates evaluation and produces a final Deny decision.
AWS Organizations enforces two levels of guardrails: Service Control Policies (SCPs) restrict the maximum permissions of principals within member accounts, while Resource Control Policies (RCPs) restrict the maximum actions permitted on resources within the organization.
In intra-account authorization, an explicit Allow in either an identity-based policy OR a resource-based policy grants access; in cross-account authorization, BOTH the identity-based policy and resource-based policy (or role trust policy) must allow the request.
AWS KMS is a critical architectural exception: a KMS key policy must explicitly allow the caller or delegate administration to the account root principal for identity-based policies to have any effect, even within the same account.
Last updated: September 2026
9.1 Complete AWS Policy Evaluation Logic & Decision Flow
Every incoming API request in AWS undergoes rigorous evaluation by the AWS authorization engine. Whether an engineer invokes s3:GetObject via the AWS Management Console, an AWS Lambda function invokes dynamodb:PutItem via an SDK, or an external automated pipeline pushes container layers to Amazon Elastic Container Registry (Amazon ECR), the request is evaluated against multiple policy types concurrently.
Securing enterprise architectures requires a comprehensive understanding of the seven-stage policy evaluation flow , the operational distinction between intra-account and cross-account requests, and the specific architectural anomalies exhibited by cryptographic services such as AWS Key Management Service (AWS KMS) .
The Complete AWS Policy Evaluation Hierarchy
When AWS receives an API request, the authorization engine evaluates the request context (the calling principal, the requested action, the target resource, and context variables like source IP, VPC endpoints, and request tags) across six discrete evaluation phases:
Loading diagram...
Stage 1: The Default State (Implicit Deny)
All AWS API requests begin in an implicit deny state. If no applicable policy explicitly grants permission for the requested action on the target resource, the authorization engine denies access by default. An implicit deny is not an active rejection statement; it is simply the absence of an explicit allow.
Stage 2: The Explicit Deny Check
The authorization engine inspects every policy that applies to the request context. This includes Service Control Policies (SCPs), Resource Control Policies (RCPs), IAM identity-based policies, resource-based policies, IAM permission boundaries, and STS session policies. If any single statement in any of these policies specifies "Effect": "Deny" matching the action, resource, and conditions, the evaluation engine terminates immediately with a final decision of DENY . An explicit deny cannot be overridden by any number of explicit allows.
Stage 3: AWS Organizations Service Control Policies (SCPs)
If the requesting principal resides in an AWS member account managed by AWS Organizations, the request must pass through the account's SCP hierarchy (evaluated from the Organization Root, through each nested Organizational Unit [OU], down to the member account). SCPs define the maximum permissions ceiling for IAM principals in member accounts. If an SCP denies the action or fails to include an allow for the service, the request is denied.
Exam Tip : SCPs do not apply to the AWS Organizations Management Account , nor do they restrict AWS Service-Linked Roles . However, SCPs apply to all member account principals, including the member account's root user.
Stage 4: AWS Organizations Resource Control Policies (RCPs)
Resource Control Policies (RCPs) enforce centralized guardrails on the resource side across accounts within an AWS Organization. While SCPs filter what principals within an organization can do, RCPs restrict the maximum permissions that can be granted on AWS resources (such as Amazon S3 buckets, AWS KMS keys, Amazon SQS queues, and AWS Secrets Manager secrets) regardless of where the requesting principal originates.
Custom RCPs contain only Deny statements; the AWS managed RCPFullAWSAccess policy, attached to every root, OU, and account when RCPs are enabled, allows everything else. A request to a resource in a member account is therefore blocked when an applicable RCP explicitly denies it. RCPs cover a defined set of services (for example Amazon S3, AWS STS, AWS KMS, Amazon SQS, and AWS Secrets Manager), don't apply to resources in the management account, and don't restrict service-linked roles.
Stage 5: Resource-Based vs. Identity-Based Policy Evaluation
Once organizational guardrails permit the action, the engine evaluates identity-based policies (attached to the IAM user, group, or role) and resource-based policies (attached to the resource, such as S3 bucket policies, KMS key policies, SQS access policies, or IAM role trust policies). The evaluation rules depend heavily on whether the request is intra-account or cross-account .
Stage 6: IAM Permission Boundaries
If the calling principal is an IAM user or IAM role that has an IAM Permission Boundary attached, the requested action must be explicitly permitted by the boundary policy. The permission boundary acts as a mathematical intersection filter: Effective Permissions = Identity-Based Policy ∩ Permission Boundary . If the boundary does not explicitly allow the action, the request is implicitly denied. Within the same account, a resource-based policy that names an IAM user ARN or a role session ARN grants access even if the boundary doesn't allow the action, but a resource-based policy that names the IAM role ARN is still limited by the boundary's implicit deny.
Stage 7: STS Session Policies
When an IAM role is assumed using the AWS Security Token Service (AWS STS) via AssumeRole , AssumeRoleWithSAML , or AssumeRoleWithWebIdentity , the caller can pass an optional inline or managed Session Policy . The session policy further scopes down the temporary credentials. The effective permissions are the intersection of the role's identity-based policies and the session policy: Effective Session Permissions = Role Policies ∩ Session Policy .
Policy Types & Governance Architecture
The following table compares the six core policy categories evaluated by the AWS authorization engine:
Policy Type Attached To Primary Purpose Can Grant Access Directly? Applies to Management Account? Primary Evaluation Role
Service Control Policy (SCP) Org Root, OU, or Member Account Enforce organizational guardrails on principals No (Ceiling only) No Restricts maximum permissions of member account principals.
Resource Control Policy (RCP) Org Root, OU, or Member Account Enforce perimeter guardrails on resources No (Deny statements only) No Restricts maximum actions permitted on organization-owned resources.
Identity-Based Policy IAM User, Group, or Role Grant permissions to principals Yes Yes (Inside account) Grants operational capabilities to AWS identities.
Resource-Based Policy AWS Resource (S3, KMS, SQS, etc.) Control access to specific resources Yes Yes Specifies which principals can access the resource directly.
Permission Boundary IAM User or Role Enforce maximum permissions ceiling on identity No (Ceiling only) Yes (If attached) Intersects with identity-based policies to prevent privilege escalation.
STS Session Policy Ephemeral STS session Scope down assumed role permissions No (Ceiling only) Yes (If passed) Narrows temporary credential permissions at session creation.
Intra-Account vs. Cross-Account Authorization Rules
One of the most frequently tested concepts on the AWS Certified Security – Specialty exam is the architectural difference between intra-account and cross-account evaluation logic.
Intra-Account Evaluation (OR Logic / Union)
When the requesting principal and the target resource reside in the same AWS account , the authorization engine evaluates identity-based policies and resource-based policies using OR logic :
An explicit Allow in either the identity-based policy OR the resource-based policy is sufficient to grant access, provided there is no explicit Deny in any applicable policy and permission boundaries permit the action.
Example : User Alice in Account 111122223333 needs to read from S3 bucket corp-data in Account 111122223333 . If Alice has an identity-based policy allowing s3:GetObject on corp-data/* , she can read objects even if the S3 bucket policy is completely empty. Conversely, if Alice has no identity policies, but the S3 bucket policy explicitly allows arn:aws:iam::111122223333:user/Alice to perform s3:GetObject , access is granted.
Cross-Account Evaluation (AND Logic / Intersection)
When the requesting principal resides in Account 111111111111 (Source Account) and the target resource resides in Account 222222222222 (Destination Account), the authorization engine enforces AND logic :
The principal's account must allow outbound access via an identity-based policy.
The resource's account must allow inbound access via a resource-based policy (or through an IAM role trust policy if using sts:AssumeRole ).
Both accounts must explicitly permit the access . If either account omits the allow, or if either account contains an explicit deny, the request is denied.
Loading diagram...
The KMS Key Policy Anomaly: Intra-Account Delegation
While most AWS services follow the intra-account OR rule, AWS KMS key policies operate under fundamentally different mechanics .
A KMS Customer Managed Key (CMK) cannot be accessed by any principal—even within the same AWS account—unless the KMS key policy explicitly grants permission. Identity-based policies in the same account have zero authority over a KMS key unless the key policy explicitly delegates control to the account root.
To enable identity-based policies to grant KMS permissions within the same account, the KMS key policy must contain the following default delegation statement: {
- "Sid": "Enable IAM User Permissions via Account Root",
- "Effect": "Allow",
- "Principal": {
- "AWS": "arn:aws:iam::111122223333:root"
- },
- "Action": "kms:*",
- "Resource": "*"
- }
In this statement, specifying the account root ( arn:aws:iam::111122223333:root ) does not grant every user admin access; rather, it delegates authorization authority to the account's IAM engine. Once this statement is present, identity-based policies in Account 111122223333 can grant kms:Decrypt , kms:GenerateDataKey , or kms:DescribeKey .
Critical Exam Trap : If a security engineer creates a custom KMS CMK and deletes the root delegation statement, an IAM administrator with AdministratorAccess ( *:* ) will receive AccessDenied when attempting to manage or decrypt with that KMS key. The only way to regain access without AWS Support is if another principal explicitly named in the key policy modifies the key policy.
Cross-Account S3 + KMS Access: The Three-Policy Triad
When an IAM principal in Account 111111111111 reads an object encrypted with SSE-KMS from an S3 bucket in Account 222222222222 , access requires a synchronized three-policy triad :
Account 111111111111 IAM Identity Policy : Must grant s3:GetObject on arn:aws:s3:::target-bucket/* AND kms:Decrypt on arn:aws:kms:region:222222222222:key/key-uuid .
Account 222222222222 S3 Bucket Policy : Must grant s3:GetObject with "Principal": {"AWS": "arn:aws:iam::111111111111:role/WorkerRole"} .
Account 222222222222 KMS Key Policy : Must grant kms:Decrypt with "Principal": {"AWS": "arn:aws:iam::111111111111:role/WorkerRole"} .
If the KMS key in Account 222222222222 uses the default AWS-managed key aws/s3 , cross-account access is impossible . AWS-managed KMS key policies cannot be modified, and aws/s3 only permits access to principals within the same account.
Organizations Resource Control Policies (RCPs) vs. Service Control Policies (SCPs)
AWS Organizations provides two distinct guardrail types that operate at opposite ends of an API transaction:
Loading diagram...
Architectural Distinctions
Principal Perimeter (SCPs) : SCPs attach to the caller's organizational hierarchy. They prevent an internal employee or compromised compute instance from making unauthorized API calls (e.g., preventing any member account principal from executing ec2:StopLogging or leaving the organization).
Resource Perimeter (RCPs) : RCPs attach to the resource's organizational hierarchy. They prevent organization-owned resources from being accessed by unauthorized external principals or unapproved networks, regardless of how permissive a resource-based policy might be configured.
Defense Against Misconfigured Bucket Policies : If a junior engineer accidentally applies an S3 bucket policy allowing "Principal": "*" without IP or condition restrictions, an active RCP that restricts S3 access strictly to corporate AWS Organizations principals prevents public data exposure.
Specialty Exam Pitfalls & Architectural Traps
The Management Account SCP Exemption : SCPs never apply to principals in the AWS Organizations management account. If an exam question describes an SCP denying s3:DeleteBucket , but an administrator in the management account is able to delete a bucket, this is expected behavior. To restrict the management account, use IAM identity-based policies or permission boundaries.
Permission Boundaries vs. Resource-Based Policies : The answer depends on which principal the resource policy names. If an S3 bucket policy in the same account names the role ARN ( arn:aws:iam::111122223333:role/AppRole ), the role's boundary still applies, so a boundary that omits Amazon S3 blocks access. If the policy names the role session ARN ( arn:aws:sts::111122223333:assumed-role/AppRole/session-name ) or an IAM user ARN , the grant is not limited by the boundary's implicit deny.
NotPrincipal with Effect: Deny : Using NotPrincipal with "Effect": "Deny" in a resource policy denies all principals in the world except the ones listed. However, this pattern can accidentally deny AWS service-linked roles, replication agents, and AWS internal callers unless paired with condition keys like aws:PrincipalArn or aws:PrincipalAccount .
STS Session Policy Limitations : A session policy can never grant permissions beyond what the underlying assumed role possesses. It can only maintain or reduce permissions. If the underlying role lacks dynamodb:Query , passing a session policy that allows dynamodb:Query results in an implicit deny.