Communication and Network Security

AI may be intelligent, but every AI system still depends on networks, protocols, APIs, trust boundaries, and communications paths.

For CISSP Domain 4 — Communication and Network Security — candidates should look beyond the model itself and ask:

An AI model can be well secured internally and still expose the organization if its network architecture allows excessive access, unrestricted outbound connections, insecure APIs, or poorly controlled communication with sensitive systems.

The Domain 4 mindset is simple:

Control the path. Limit the trust. Protect the communication.

Why AI Security Matters in CISSP Domain 4

Modern AI rarely operates as a standalone system.

A typical enterprise AI service may communicate with:

• users;

• web applications;

• APIs;

• databases;

• vector databases;

• cloud services;

• identity providers;

• plugins;

• AI agents;

• external models;

• internal applications; and

• security monitoring systems.

Each connection creates a potential communication path and attack surface.

For CISSP candidates, AI security therefore extends naturally into:

segmentation, secure protocols, network access control, encryption in transit, API security, Zero Trust, monitoring, availability, and secure network architecture.

1. Network Segmentation Limits AI Risk

AI systems may process untrusted prompts while also accessing sensitive enterprise resources.

Putting everything on the same trusted network dramatically increases risk.

Segmentation can separate:

• public-facing AI services;

• model infrastructure;

• sensitive datasets;

• development environments;

• production environments;

• administrative systems;

• AI agents; and

• critical business applications.

If an AI component is compromised, segmentation can reduce the attacker's ability to move laterally.

CISSP Exam Thinking

When an AI service does not require unrestricted access to the enterprise network, do not give it unrestricted access.

This reflects both:

least privilege and defense in depth.

2. Zero Trust Applies Naturally to AI

AI systems should not receive implicit trust merely because they operate inside the corporate network.

Zero Trust principles remain highly relevant:

Never trust automatically. Verify explicitly. Limit access continuously.

An AI workload requesting access to an internal resource should be evaluated based on factors such as:

• identity;

• authorization;

• device or workload context;

• requested resource;

• sensitivity;

• policy; and

• risk.

CISSP Insight

An internally hosted AI system can still be compromised, manipulated, misconfigured, or abused.

3. AI APIs Are Critical Attack Surfaces

AI services frequently communicate through APIs.

These APIs may allow users or applications to:

• submit prompts;

• retrieve information;

• invoke models;

• upload files;

• execute tools;

• retrieve embeddings;

• access organizational data; or

• trigger automated actions.

Weak API security can expose the entire AI environment.

Important protections include:

• strong authentication;

• authorization;

• encrypted communications;

• input validation;

• rate limiting;

• logging;

• secure token handling; and

• restricted API scopes.

CISSP Principle

Authentication proves identity. Authorization determines what that identity may do.

An authenticated AI client should never automatically receive unrestricted API access.

4. Protect AI Communications in Transit

AI traffic may contain extremely sensitive information:

• prompts;

• customer data;

• proprietary documents;

• credentials;

• model responses;

• source code;

• security findings; and

• business information.

Communications should therefore receive protection appropriate to their sensitivity.

Encryption in transit can protect against:

• interception;

• unauthorized disclosure; and

• certain forms of tampering.

Protocols such as TLS are commonly used to protect communications between:

User ↔ Application ↔ API ↔ AI Service ↔ Data Source

Exam Trap

Encryption protects the communication channel.

It does not determine whether the communicating party should have access in the first place.

CISSP candidates must distinguish:

secure communication from authorized communication.

You often need both.

5. Control Outbound AI Communication

Inbound protection receives considerable attention, but AI systems may also initiate outbound connections.

An AI agent could potentially communicate with:

• external APIs;

• websites;

• cloud services;

• plugins;

• third-party tools; or

• attacker-controlled infrastructure.

Unrestricted outbound access creates opportunities for:

• data exfiltration;

• malicious tool execution;

• command-and-control communication;

• unauthorized information sharing; and

• circumvention of security boundaries.

Architectural controls may restrict AI workloads to approved destinations and required protocols.

CISSP Insight

If an AI system needs access to three services, it should not automatically receive access to the entire Internet.

6. AI Agents Increase Network Risk

Traditional AI chat systems generate responses.

AI agents may take actions.

An agent might:

• query databases;

• modify cloud resources;

• send messages;

• create tickets;

• execute scripts;

• access APIs; or

• interact with production systems.

This dramatically increases the importance of network controls.

An AI agent compromised through malicious input could potentially use its legitimate network connections to reach other systems.

Therefore, AI agents should have:

• narrowly scoped connectivity;

• strong service identities;

• restricted API permissions;

• segmented environments;

• monitored communication; and

• approval requirements for high-impact actions.

CISSP Exam Thinking

Do not secure only the AI agent.

Secure what the agent is allowed to reach.

7. RAG Changes the Network Trust Model

Retrieval-Augmented Generation (RAG) allows an AI system to retrieve external or internal information before generating an answer.

A simplified communication flow might be:

User → AI Application → Retrieval Service → Vector Database → Enterprise Data → Model

Every communication path creates a trust relationship.

Security questions include:

• Is communication encrypted?

• Is the requesting identity authenticated?

• Is access authorized?

• Can the AI retrieve information outside the user's permissions?

• Are retrieval systems segmented?

• Are queries and responses logged?

• Can untrusted documents influence downstream behavior?

CISSP Principle

The AI layer must not bypass network and access controls protecting the underlying data.

8. DNS and Name Resolution Still Matter

AI applications rely on traditional infrastructure.

If DNS is manipulated, an AI workload could potentially be redirected toward malicious infrastructure.

Threats involving name resolution may contribute to:

• traffic redirection;

• credential theft;

• interception;

• malicious API endpoints; or

• communication with attacker-controlled services.

AI does not eliminate foundational network-security concepts.

CISSP Takeaway

When troubleshooting an AI communication problem on the exam, do not become distracted by the word AI.

The underlying problem may still be:

DNS, routing, TLS, firewall policy, segmentation, authentication, or network availability.

9. Availability Matters for AI Services

Organizations may increasingly depend on AI for:

• customer service;

• fraud detection;

• security analysis;

• operational decisions;

• software development;

• automation; and

• business workflows.

This makes availability a security concern.

Threats include:

• denial-of-service attacks;

• API exhaustion;

• resource exhaustion;

• dependency failures;

• cloud outages; and

• excessive or malicious requests.

Controls may include:

• rate limiting;

• redundancy;

• resilient architecture;

• capacity planning;

• traffic filtering;

• load balancing; and

• tested recovery mechanisms.

CISSP Insight

An AI system that cannot support required business operations has failed an availability objective, regardless of model accuracy.

10. Rate Limiting Is Especially Important for AI

AI requests can consume substantial computational resources.

Attackers—or poorly designed applications—may generate excessive requests that consume:

• API quotas;

• GPU capacity;

• memory;

• bandwidth; or

• financial resources.

Rate limiting can reduce abuse and help preserve availability.

It may also help mitigate:

• automated attacks;

• resource exhaustion;

• excessive API use; and

• certain model-extraction attempts.

But remember:

Rate limiting is one layer, not a complete security architecture.

11. Monitor AI Network Traffic

AI workloads should not become invisible parts of the network.

Organizations should understand:

• which systems communicate with AI services;

• what destinations AI workloads contact;

• unusual traffic patterns;

• excessive API requests;

• unauthorized connections;

• failed authentication attempts; and

• unexpected data transfers.

Useful controls may include:

• centralized logging;

• network monitoring;

• IDS/IPS;

• anomaly detection;

• SIEM integration; and

• traffic analysis.

CISSP Exam Thinking

Monitoring provides visibility and detection.

It does not replace preventive controls such as segmentation, authentication, authorization, and encryption.

12. Network Security Controls Must Follow Risk

Not every AI workload requires the same architecture.

A public chatbot answering general questions has a different risk profile from an AI agent capable of modifying production infrastructure.

Security architecture should therefore reflect:

• asset sensitivity;

• business impact;

• required connectivity;

• threat exposure;

• privileges;

• regulatory requirements; and

• consequences of compromise.

The more powerful the AI system becomes, the more important restricted communication paths and strong trust boundaries become.

The CISSP AI Network Security Mental Model

When an AI-related Domain 4 question appears, think:

Identify the Communication Path → Establish Trust Boundaries → Authenticate → Authorize → Segment → Encrypt → Restrict → Monitor → Maintain Availability

Ask:

Who is communicating?

What systems can they reach?

Is the communication authorized?

Is the channel protected?

Can unnecessary connections be removed?

What happens if the AI component is compromised?

That is the Domain 4 perspective.

Example CISSP Question

An organization deploys an AI agent that analyzes internal security incidents. The agent requires access to a threat-intelligence API, a security database, and a ticketing platform. Security discovers that the AI server currently has unrestricted outbound Internet access.

What is the BEST security improvement?

A. Increase the AI model's training dataB. Encrypt the model's storage volumeC. Restrict network communication to authorized destinations and required servicesD. Increase the AI server's processing capacity

Best answer: C

The agent has defined communication requirements.

Allowing unrestricted outbound connectivity violates the principle of least privilege and increases the potential for data exfiltration or unauthorized communication.

Encryption protects stored information but does not address unnecessary network access.

Exam Thinking

The best CISSP answer usually addresses the risk directly and proportionately.

If the system only needs specific communication paths:

High-Yield AI Security Rules for CISSP Domain 4

Remember these principles:

1. AI systems still depend on traditional network-security fundamentals.

2. Segment AI workloads according to risk and trust boundaries.

3. Do not implicitly trust AI systems because they are inside the corporate network.

4. Secure AI APIs with authentication and authorization.

5. Protect sensitive AI communications in transit.

6. Encryption does not replace access control.

7. Restrict unnecessary outbound AI connectivity.

8. Apply network least privilege to AI agents.

9. RAG systems must preserve underlying access and network boundaries.

10. Monitor AI network communication for abnormal behavior.

11. Protect AI services against resource exhaustion and availability attacks.

12. Do not let the word “AI” distract you from DNS, routing, segmentation, firewalls, TLS, APIs, and other core network principles.

Final CISSP Takeaway

Domain 4 teaches an important AI security lesson:

An AI system is only as secure as the communication paths connecting it to the rest of the environment.

When you see AI in a CISSP network-security scenario, do not immediately search for an exotic AI-specific control.

First ask:

What is connected? What should be connected? Who should communicate? What should they be allowed to reach? How is the communication protected?

Then apply familiar CISSP principles:

segmentation, Zero Trust, least privilege, secure protocols, API security, monitoring, resilience, and defense in depth.

AI may change what travels across the network. It does not change the need to control who can communicate with what.