# Network Isolation

Network isolation separates resources and restricts their communication to prevent unauthorized access and limit the spread of security incidents.

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

> Network isolation creates controlled communication boundaries between workloads, environments, and external networks. During incident response, it helps contain compromised resources, interrupt attacker communications, and protect unaffected systems while engineers investigate, eradicate threats, and restore trusted workloads.

Practice exams covering Network Isolation: [ANS-C01 AWS Certified Advanced Networking - Specialty](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty), [AIF-C01 AWS Certified AI Practitioner](https://blowstack.com/practice-exams/aws-certified-ai-practitioner), [SOA-C03 AWS Certified CloudOps Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate), [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).

## 6 key aspects of Network Isolation

Effective network isolation combines explicit boundaries, bidirectional restrictions, evidence preservation, and validated recovery without confusing containment with threat removal.

1. **Explicit communication boundaries**: Define which resources may communicate, over which protocols, and for what purpose. Include private routes, shared networks, and service endpoints rather than considering only internet exposure.
2. **Bidirectional containment**: Restrict outbound communication as well as inbound access to limit exfiltration, command-and-control traffic, and lateral movement. Preserve investigation access only when its risk is understood and accepted.
3. **Control semantics and scope**: Choose controls according to their statefulness, deny capabilities, and enforcement scope. Account for tracked connections and the collateral impact of subnet-wide or shared-firewall changes.
4. **Evidence-aware response**: Contain active harm while preserving useful logs, configuration records, and forensic evidence. Avoid unnecessary destructive actions that could prevent determination of the incident's cause and extent.
5. **Separation from eradication**: Treat isolation as containment, not proof that a resource is clean. Remove persistence, remediate vulnerabilities, and address compromised credentials before recovery.
6. **Validated restoration**: Restore trusted workloads into a restricted environment and verify them before reconnecting production traffic. Exercise containment and recovery runbooks so that emergency changes have predictable results.

## Cloud services for Network Isolation

Amazon VPC supplies the primary isolation boundaries and traffic controls, while AWS Network Firewall adds inspection for traffic deliberately routed through it. For compromised EC2 workloads, responders can restrict network interfaces and preserve storage evidence before rebuilding trusted replacements. AWS Systems Manager can support controlled investigation and repeatable response actions when its management connectivity remains available. These services must be configured as a coordinated containment design: restrictive rules do not help if another interface, route, or permissive security group leaves an unintended path open. Recovery also requires threat eradication, credential remediation, and validation before normal connectivity is restored.

- **[Amazon VPC](https://blowstack.com/tools/cloud-services-hub/amazon-vpc)**: Amazon VPC provides virtual network boundaries, route tables, security groups, and network ACLs that support isolation. Security groups control traffic for associated network interfaces, while subnet-level network ACLs provide stateless allow and deny rules.
- **[AWS Network Firewall](https://blowstack.com/tools/cloud-services-hub/aws-network-firewall)**: AWS Network Firewall applies stateful and stateless inspection rules to traffic routed through its endpoints. It can enforce containment policies for relevant network paths, but traffic that bypasses those endpoints is not inspected.
- **[Amazon EC2](https://blowstack.com/tools/cloud-services-hub/amazon-ec2)**: For affected EC2 instances, engineers can replace security group associations, examine attached network interfaces, and create snapshots of EBS volumes for investigation. Rebuilding replacement instances in a clean, restricted network supports recovery without reconnecting the compromised operating system.
- **[AWS Systems Manager](https://blowstack.com/tools/cloud-services-hub/aws-systems-manager)**: Systems Manager can provide managed investigation access and execute remediation or containment automation without opening inbound administration ports. Its use during quarantine depends on the required agent, permissions, and connectivity to service endpoints.

## Network Isolation exam questions

Real practice exam questions on Network Isolation, with the correct answer and its explanation.

### Question 1 (AIF-C01)

**Question:**

A financial services company runs an application in private subnets in an Amazon VPC. The application uses Amazon Bedrock to summarize confidential reports. Its IAM role already permits the required model invocations. Company policy prohibits internet gateways and NAT gateways for this VPC and requires inference traffic to remain on the AWS network. Which solution BEST meets these requirements?

**Correct answer:** Create an AWS PrivateLink interface VPC endpoint for the Amazon Bedrock runtime, enable private DNS, and allow HTTPS access from the application.

**Explanation:** An interface VPC endpoint for the Bedrock runtime provides private connectivity for model inference without an internet gateway or NAT gateway. Private DNS allows the application to resolve the service hostname to the endpoint's private IP addresses.

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

**Question:**

A security team detects that an Amazon EC2 instance used for ML model evaluation is sending proprietary data to an external IP address over an established TCP connection. The instance also has established TCP connections to application resources in other subnets of the same VPC. It is the only resource in its subnet, has one network interface, and uses IPv4 only.

The team must immediately block traffic in both directions between the instance and all external IP addresses and application resources in other subnets, including traffic on established connections. The instance must remain running for forensic analysis, and other subnets must remain unaffected. Traffic to AWS-provided infrastructure services, such as AmazonProvidedDNS, is outside the containment requirement.

Which action meets these requirements?

**Correct answer:** Associate the instance's subnet with a dedicated custom network ACL that has no inbound or outbound allow rules.

**Explanation:** Network ACLs are stateless, so their default deny rules block packets in both directions, including packets belonging to established connections. This blocks the specified external and inter-subnet traffic. Because the subnet contains only this instance and the ACL is dedicated to that subnet, other subnets remain unaffected.

### Question 3 (AIP-C01)

**Question:**

A research company deploys a foundation model on an Amazon SageMaker AI real-time endpoint by using a third-party inference container. The model artifacts are stored in Amazon S3. The container loads the complete model at startup and does not need to retrieve additional data during inference. Application clients already invoke the endpoint through a private connection.

A security review identifies a risk that code in the inference container could transmit proprietary model weights to another destination. The company requires the container to be unable to initiate network connections, including connections to AWS APIs. SageMaker AI must still retrieve the deployment artifacts, deliver inference requests, and collect the container's standard output logs.

Which deployment configuration meets these requirements?

**Correct answer:** Set EnableNetworkIsolation to true when creating the SageMaker AI model. Grant the SageMaker AI execution role the permissions required for platform-managed artifact retrieval and log delivery, without relying on container-initiated API calls.

**Explanation:** SageMaker AI network isolation prevents the inference container from making outbound network calls, including calls to AWS services, and does not expose AWS credentials to the container. SageMaker AI performs artifact retrieval and other required hosting operations separately from the isolated container, while continuing to deliver inference requests and collect standard output logs.

### Question 4 (SOA-C03)

**Question:**

A company runs an analytics workload in an isolated VPC. The VPC has an Amazon S3 gateway endpoint and no internet gateway, NAT gateway, peering connection, or transit gateway attachment. The endpoint uses its default full-access policy. The workload's IAM role permits reading several S3 buckets, but workloads in this VPC must be able to read only the company-analytics-input bucket. The same IAM role is used in other VPCs where access to the other buckets must remain available.

Testing shows that the isolated workload can read objects from the other buckets through the gateway endpoint. Which change should a CloudOps engineer make to enforce the required boundary?

**Correct answer:** Replace the gateway endpoint policy with a policy that allows only s3:ListBucket on company-analytics-input and s3:GetObject on objects in that bucket.

**Explanation:** A gateway endpoint policy limits access through that endpoint. Replacing the full-access policy with these resource-scoped permissions leaves requests to other buckets implicitly denied through this VPC's only S3 access path. The IAM role's permissions in other VPCs remain unchanged.

### Question 5 (SCS-C03)

**Question:**

After containing a malware incident, a company is automating recovery of an Amazon EC2 application server. The workflow launches a replacement from a verified clean AMI and restores application data into a quarantine environment. The replacement initially has no connectivity to production.

A validation script checks the restored data and application configuration. The script completes successfully even when validation fails and returns a Boolean output named ValidationPassed. Production connectivity must be enabled only if this output is true and an incident commander subsequently approves the recovery. A validation error, failed validation, approval rejection, or approval timeout must leave the replacement quarantined.

Which AWS Systems Manager Automation runbook design meets these requirements?

**Correct answer:** Run validation with aws:executeScript, and then use aws:branch to route only ValidationPassed=true to an aws:approve step. Route false and default outcomes to an end step that retains quarantine. Configure validation and approval errors to abort execution, and enable production connectivity only after approval succeeds.

**Explanation:** The branch explicitly evaluates the validation result rather than treating successful script execution as proof that validation passed. Only a true result reaches the subsequent approval gate. False or unexpected results end without releasing quarantine, and aborting on validation or approval errors prevents rejection, timeout, or execution failure from reaching the connectivity step.

## Practice Network Isolation in a full exam

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

- [SOA-C03](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate): 5 questions
- [SCS-C03](https://blowstack.com/practice-exams/aws-certified-security-specialty): 5 questions
- [AIP-C01](https://blowstack.com/practice-exams/aws-certified-generative-ai-developer-professional): 4 questions
- [MLA-C02 (beta)](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate): 3 questions
- [AIF-C01](https://blowstack.com/practice-exams/aws-certified-ai-practitioner): 2 questions
- [ANS-C01](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty): 2 questions

## What is Network Isolation

Network isolation is the deliberate restriction of network communication between resources or groups of resources. The objective is not always complete disconnection: an isolated workload may retain narrowly authorized access to forensic tools, logging systems, or management endpoints. The essential requirement is that only explicitly justified communication remains possible. Isolation can be a permanent architectural boundary or a temporary containment measure applied during a security incident.

In cloud environments, engineers implement isolation using virtual networks, subnet boundaries, routing controls, workload firewalls, network access control lists, and inspection gateways. Separate virtual networks provide useful boundaries, but peering connections, transit networks, private endpoints, and shared services can create additional paths. A resource without a public IP address is not necessarily isolated: it may still reach the internet through a NAT gateway or communicate with compromised systems over private routes.

During an incident, containment should address both inbound and outbound traffic. Blocking inbound connections alone does not prevent malware from contacting command-and-control infrastructure, exfiltrating data, or attacking neighboring workloads. For example, an engineer investigating a compromised application server can remove it from a load balancer, replace its existing security groups with a restrictive quarantine group, and preserve only approved investigation access. All attached network interfaces and remaining access paths must be considered; adding a restrictive security group alongside permissive groups does not override their allowances.

Control behavior matters when selecting a containment action. AWS security groups are stateful, allow-only controls, and changing their rules does not necessarily terminate tracked connections immediately. Network ACLs are stateless subnet-level controls with explicit deny rules and can interrupt matching traffic, including existing flows. However, an ACL change can affect every resource in the associated subnet, and legitimate connections require appropriate rules in both directions. Some platform traffic, including access to the Amazon-provided DNS resolver, is not filtered by security groups or network ACLs, so these controls should not be treated as a universal packet-blocking mechanism.

Preserve evidence while limiting ongoing harm. Record resource identifiers, network configuration, relevant logs, and the containment actions taken. Where appropriate, capture disk snapshots and volatile evidence before stopping or rebuilding a resource, provided collection does not unacceptably delay containment. An isolated network does not revoke stolen credentials or remove malware. Investigators must also address compromised identities, persistence mechanisms, vulnerable software, and any other affected resources.

Plan investigation access before an emergency. A private management path can avoid reopening public SSH or RDP, but management tools still need working connectivity, permissions, and often an agent. For example, a managed session may depend on interface endpoints or controlled outbound access. Complete disconnection can prevent evidence collection, logging, health checks, and remote remediation. The appropriate balance depends on the severity of the threat and the trustworthiness of the remaining management path.

Recovery should restore trust rather than merely restore connectivity. Rebuild affected workloads from trusted images or restore validated backups into a clean, restricted environment. Patch the original weakness, rotate exposed credentials, check for persistence, and verify application behavior before gradually reconnecting production traffic. Restoring a backup without checking when compromise began can reintroduce the threat. Test isolation runbooks in advance and verify containment with traffic observations and connectivity tests rather than assuming that a successful configuration change proves the resource is contained.

## Learn Network Isolation in lessons

6 short lessons on Network Isolation in 5 certification courses.

- AIF-C01: Network Isolation: private connectivity for AI workloads (Premium), https://app.blowstack.com/lessons/network-isolation-private-connectivity-for-ai-workloads-aif-c01
- MLA-C02 (beta): Network Isolation vs. Network Segmentation (Premium), https://app.blowstack.com/lessons/network-isolation-vs-network-segmentation-mla-c02
- AIP-C01: Network Isolation: private communication boundaries for AI workloads (Premium), https://app.blowstack.com/lessons/network-isolation-private-communication-boundaries-for-ai-workloads-aip-c01
- SOA-C03: Network Isolation: boundaries for private connectivity (Premium), https://app.blowstack.com/lessons/network-isolation-boundaries-for-private-connectivity-soa-c03
- SCS-C03: Network Isolation: incident quarantine boundaries (Premium), https://app.blowstack.com/lessons/network-isolation-incident-quarantine-boundaries-scs-c03
- MLA-C02 (beta): Network Isolation: communication boundaries for ML workloads (Premium), https://app.blowstack.com/lessons/network-isolation-communication-boundaries-for-ml-workloads-mla-c02
