Amazon AWS Security: Shared Responsibility and IAM

AWS offers managed infrastructure and security services, but a workload is not automatically secure because it runs on AWS. The central operating principle is shared responsibility: AWS secures the infrastructure that supports its cloud services, while customers remain responsible for the data, identities, settings, applications, and other controls within their scope. The exact boundary changes with the service.

For an administrator, that principle becomes concrete through AWS Identity and Access Management (IAM), network boundaries, encryption, monitoring, and tested response procedures. This guide connects those controls to realistic decisions instead of treating security as a checklist of products.

Understand who owns each layer of security

AWS is responsible for security of the cloud: facilities, physical equipment, and underlying infrastructure operated by AWS. A customer is responsible for appropriate security in the cloud, including who can access information and what configurations the team chooses. This is not a blanket division in which AWS fixes every operating-system or application problem.

Consider EC2. The customer normally manages guest operating-system configuration and patching, application software, IAM permissions, and controls such as security groups. For a managed service such as Amazon S3 or DynamoDB, AWS operates additional infrastructure and software layers, while the customer still owns data classification, authorization, encryption options, and service configuration.

Write a responsibility matrix for one actual workload. For each component, record which organization manages the infrastructure, operating system, application, identities, data, and recovery plan. Repeat the matrix when the team moves a database or application from self-managed infrastructure to a managed service; the duties will not remain identical.

Use IAM roles and temporary credentials deliberately

IAM policies determine which principals may perform particular actions on resources and under which conditions. Effective permissions emerge from identity policies, resource policies, explicit denies, organizational guardrails, and other relevant controls. A broad allow in one location cannot override an applicable explicit deny elsewhere.

For human access, AWS recommends federation through an identity provider, commonly managed with IAM Identity Center, so users can obtain temporary credentials. Protect root-user access, require strong multifactor authentication, and avoid using root credentials for everyday administration. Applications should generally use IAM roles with temporary credentials rather than embedding long-lived access keys.

Least privilege begins with a defined task: identify exactly what an application needs, grant an appropriately scoped policy, and verify both successful and denied operations. An application that only reads a specific S3 path should not automatically receive write or delete access to every bucket in an account.

IAM Access Analyzer can help identify unintended public or cross-account access and review policy intent. Revisit dormant permissions and roles as the workload changes. A permission that was justified during migration may become a risk once the migration is complete.

Protect data through access, encryption, and lifecycle controls

Start by classifying what the workload stores. Sensitive customer data, public assets, audit records, and temporary build artifacts require different access and retention policies. A secured network does not compensate for a resource policy that unintentionally grants public access.

For S3, evaluate account- and bucket-level public-access controls, identity and bucket policies, object ownership, and supported encryption settings. For databases, review authentication, network exposure, encryption, audit logs, and backup permissions. Check where copies of data exist in snapshots, exports, development accounts, and logs.

AWS Key Management Service helps manage cryptographic keys used by supported AWS services. It does not, by itself, establish all encryption-in-transit behavior or guarantee that only authorized application users can read data. Our KMS key management and encryption controls guide provides additional context for keys, secrets, and certificates. Keep policy and key access aligned with the application trust boundary.

Test access with two roles: one that should read a protected object and another that should not. Verify that encryption is enabled as intended, that denied access is logged where configured, and that restoration from a backup does not silently expose sensitive data to a broader audience.

Network controls are only one part of the boundary

Within a VPC, route tables establish the network path, security groups control stateful traffic at supported resources, and network ACLs provide stateless subnet-level rules. Private subnets, gateways, VPC endpoints, and load balancers affect which paths exist. None of these features can replace correct application authorization.

When troubleshooting a connection, first determine the expected source, destination, port, DNS resolution, and route. Check the actual path before changing security rules. Opening an unrestricted inbound port may make a test pass while creating a larger security exposure.

For administrative access, avoid assuming that direct internet exposure is necessary. Evaluate approved private access, Systems Manager capabilities, and identity-based controls as appropriate to the workload. Verify both an allowed administration path and a denied unauthorized one.

Know what the main AWS security services actually detect

Different services produce different evidence. Amazon GuardDuty analyzes supported telemetry for potentially malicious or unauthorized behavior. Amazon Inspector assesses supported compute and container resources for relevant vulnerabilities and exposure information. Amazon Macie focuses on sensitive-data discovery in supported S3 environments. They should not be described as interchangeable scanners.

AWS Security Hub and AWS Security Hub CSPM are complementary but distinct. Security Hub CSPM evaluates configurations against security controls and standards; the enhanced Security Hub correlates findings and supports prioritization and response across enabled sources. A central dashboard does not mean every required service is enabled, every account is covered, or every finding is a confirmed incident.

AWS WAF helps enforce web-request rules on supported applications, while AWS Shield provides DDoS-protection capabilities. AWS Config can record and assess supported resource configurations. Choose controls based on the attack or misconfiguration you need to prevent or detect, and document their coverage gaps.

Review one synthetic finding in a nonproduction environment. Identify the source, affected resource, evidence, severity, owner, and next action. A false positive, a configuration issue, and a confirmed compromise deserve different handling even if they appear in a consolidated console.

Collect evidence and prepare an incident response

CloudTrail can record important API activity, while CloudWatch metrics and logs help explain service and application behavior. Configure the necessary logging and retention for the environment; not every data event, application log, or cross-account action is automatically recorded with the same level of detail.

Design response procedures before a real incident. A runbook should describe initial triage, evidence preservation, authorization for containment, communication, and the checks required before service restoration. An automatic remediation rule can be useful for a low-risk, well-defined configuration change but harmful when it removes a critical resource during an outage.

Practise a tabletop or isolated simulation in which a role gains an excessive permission or a storage configuration violates policy. Identify how the problem would be detected, which access change is safest, how operators would validate it, and how they would avoid destroying useful audit evidence.

Use governance and continuous review, not a one-time checklist

In a multi-account organization, guardrails help prevent recurring risky configurations, but they still require appropriate design and operational ownership. Service control policies can restrict actions within an organization; they do not automatically grant users permissions. Configuration controls and security posture assessments provide complementary evidence.

Compliance obligations vary by industry, geography, data type, and contract. AWS’s provider certifications do not automatically establish that a particular customer workload satisfies all applicable requirements. Map the requirement to the controls your organization must configure, monitor, test, and document.

Review access, logs, unused resources, software updates, encryption settings, backups, and incident lessons regularly. Cloud environments change quickly; a policy that was appropriate for a prototype may become unsafe when real users, data, and integrations are introduced.

Build one practical security verification project

Choose a small nonproduction workload with an application role, private data, monitored network traffic, and a basic recovery procedure. Write down the intended trust boundaries, configure the controls, and test an allowed and denied request. Then introduce a safe misconfiguration and identify the evidence that would expose it.

Make your final report specific: what AWS manages, what you manage, which permission or network decision prevented an unauthorized action, which finding or log supported your diagnosis, and how you confirmed the correction. The objective is not to turn on every security product. It is to demonstrate accountable, measurable protection for a real workload.

  • img