Identity and Access Management

AI changes how systems interact with users, data, applications, and each other—but it does not change a fundamental security rule:

For CISSP Domain 5 — Identity and Access Management (IAM) — AI introduces an important new challenge: identity is no longer only about human users.

Modern environments may include AI applications, models, agents, APIs, service accounts, workloads, plugins, and autonomous processes, all requesting access to organizational resources.

The CISSP question becomes:

Why AI Security Matters in CISSP Domain 5

AI systems may interact with:

• users;

• identity providers;

• enterprise applications;

• APIs;

• databases;

• cloud resources;

• sensitive documents;

• email systems;

• source-code repositories;

• security tools; and

• other AI agents.

Each interaction creates an identity and authorization decision.

The core IAM principles remain:

Identification → Authentication → Authorization → Accountability

AI does not replace this sequence.

It makes enforcing it more important.

1. AI Agents Need Identities Too

Traditional IAM focuses heavily on human identities.

AI introduces powerful non-human identities.

An AI agent may need to:

• query a database;

• retrieve documents;

• create tickets;

• send messages;

• modify cloud resources;

• execute workflows; or

• call external APIs.

The agent should therefore have a unique, managed identity rather than borrowing a user's credentials or sharing an administrative account.

CISSP Principle

If multiple AI agents share the same credentials, attribution and control become significantly weaker.

2. Least Privilege Is Critical for AI

AI systems can make mistakes.

They can also be manipulated.

Giving an AI agent excessive privileges turns an incorrect or maliciously influenced decision into a potentially serious security incident.

An AI agent that only needs to read security alerts should not automatically be allowed to:

• delete accounts;

• modify firewall rules;

• disable security controls;

• change permissions; or

• execute administrative commands.

The correct approach is familiar:

CISSP Exam Thinking

When the choice is between convenience and least privilege, CISSP usually favors least privilege when business requirements can still be met.

3. Authentication Does Not Equal Authorization

This distinction becomes particularly important with AI.

An authenticated AI agent has proven—or presented evidence of—its identity.

That does not mean it should have unrestricted access.

For example:

An AI assistant may successfully authenticate to a corporate document repository.

That does not mean it should retrieve every document in the repository.

Authorization must still determine:

What resources may this identity access?

What actions may it perform?

CISSP Rule

Authentication answers: Who are you?

Authorization answers: What are you allowed to do?

AI does not change that distinction.

4. AI Must Preserve the User's Authorization Boundary

Consider an enterprise AI assistant connected to internal documents.

Employee A is authorized to access:

Documents 1, 2, and 3

Employee B is authorized to access:

Documents 4, 5, and 6

The AI assistant should not retrieve Documents 4–6 for Employee A simply because the AI itself has access to the entire repository.

This is a critical architectural and IAM principle.

The effective access should reflect the permissions of the user and the legitimate business purpose.

5. RAG Systems Create an Authorization Challenge

Retrieval-Augmented Generation (RAG) systems retrieve information before generating an answer.

A simplified flow might be:

User → AI Assistant → Retrieval System → Enterprise Data → Model → User

The danger is obvious:

If the retrieval layer ignores the user's authorization, the AI may expose information the user could never access directly.

Security should therefore enforce access controls before sensitive information is retrieved or disclosed.

CISSP Exam Thinking

Do not rely on the AI model to decide whether confidential information should be revealed.

Authorization should be enforced by trusted access-control mechanisms.

6. Service Accounts and AI Credentials Must Be Protected

AI workloads frequently use:

• API keys;

• service accounts;

• access tokens;

• certificates;

• OAuth tokens;

• cloud identities; and

• application secrets.

These credentials can become powerful targets.

If compromised, an attacker may gain the privileges assigned to the AI workload.

Controls may include:

• secrets management;

• credential rotation;

• short-lived tokens;

• restricted scopes;

• secure storage;

• monitoring; and

• elimination of hard-coded credentials.

CISSP Insight

The security of an AI agent depends partly on the security of the identity under which it operates.

7. Avoid Shared and Embedded Credentials

Hard-coding credentials into:

• prompts;

• source code;

• configuration files;

• AI workflows; or

• scripts

creates unnecessary risk.

Likewise, sharing one powerful service account among many AI workloads makes accountability and privilege management difficult.

Prefer:

unique identity → minimum privilege → protected credentials → auditable activity

This supports both security and accountability.

8. Privileged AI Requires Stronger Controls

An AI system with administrative capabilities presents substantially more risk than an AI system that simply summarizes documents.

Privileged AI actions might include:

• creating accounts;

• modifying permissions;

• changing security configurations;

• deploying software;

• deleting resources;

• resetting credentials; or

• modifying production systems.

Such capabilities should receive controls appropriate to their impact.

These may include:

• privileged access management;

• just-in-time access;

• time-limited privileges;

• approval workflows;

• session monitoring;

• stronger authentication; and

• detailed logging.

CISSP Principle

9. Human Approval Can Strengthen AI Authorization

Not every action an AI agent proposes should be executed automatically.

For low-impact activities, automation may be appropriate.

For high-impact activities, authorization architecture can require human approval.

For example:

AI detects suspicious account → recommends disabling account → authorized human approves → action occurs

This is safer than:

AI detects suspicious account → automatically disables account

when the business impact of a false positive could be significant.

CISSP Insight

AI autonomy should not exceed the organization's tolerance for the consequences of an incorrect decision.

10. Separation of Duties Still Applies

An AI system should not necessarily be able to:

request → approve → execute → verify

the same sensitive transaction.

That concentrates too much control in one entity.

For critical processes, organizations may separate responsibilities.

For example:

AI analyzes → human approves → separate system executes → monitoring verifies

This reduces opportunities for fraud, error, and abuse.

Exam Thinking

When one identity or system controls every stage of a critical process, think:

separation of duties.

11. AI Can Strengthen IAM—and Attack It

AI is not only an IAM risk.

It can also support identity security by helping detect:

• unusual login behavior;

• impossible travel;

• credential abuse;

• anomalous access;

• privilege escalation;

• compromised accounts; and

• suspicious access patterns.

But AI-generated risk scores should not automatically be treated as infallible.

Organizations must consider:

• false positives;

• false negatives;

• model manipulation;

• bias;

• data quality; and

• business impact.

CISSP Perspective

AI can improve detection.

It does not eliminate the need for sound IAM controls.

12. Deepfakes Increase Authentication Risk

Generative AI can create increasingly convincing:

• voices;

• images;

• videos; and

• impersonations.

This weakens processes that depend heavily on a single human characteristic or informal recognition.

For example, hearing an executive's voice on a phone call should not automatically authorize a sensitive financial transaction.

High-risk processes may require stronger independent verification.

CISSP Lesson

The higher the risk, the stronger the authentication requirement should be.

13. MFA Remains Important

AI-enabled phishing, social engineering, and impersonation increase the value of strong authentication.

Multi-factor authentication requires evidence from different authentication-factor categories, such as:

• something you know;

• something you have; and

• something you are.

However, MFA itself should be implemented appropriately.

Not all authentication methods provide the same resistance to phishing or impersonation.

CISSP Exam Thinking

Select authentication strength based on:

risk, sensitivity, business impact, and assurance requirements.

Do not simply choose the most complicated mechanism.

14. Access Reviews Must Include AI Identities

Permissions accumulate over time.

An AI workload that once required broad access may no longer need it.

Organizations should periodically review:

• AI service accounts;

• agent permissions;

• API scopes;

• privileged access;

• inactive identities;

• external integrations; and

• connected applications.

Ask:

Does this AI identity still need this access?

If not, remove it.

CISSP Principle

15. Logging Creates Accountability

Organizations should be able to determine:

• which user invoked the AI;

• which AI identity performed an action;

• what resource was accessed;

• what privilege was used;

• when the activity occurred;

• whether approval was obtained; and

• what action resulted.

Without sufficient logging, organizations may struggle to distinguish:

human action → AI recommendation → AI execution → attacker activity

Accountability becomes particularly important as AI systems gain greater autonomy.

The CISSP AI-IAM Mental Model

When an AI-related Domain 5 question appears, think:

Identify → Authenticate → Authorize → Least Privilege → Separate Duties → Monitor → Review → Revoke

Then ask:

Who is requesting access?

Is it a human or non-human identity?

How was the identity authenticated?

What should it be authorized to access?

Does it need this privilege?

Can a sensitive action require independent approval?

Can the activity be attributed and audited?

That is the Domain 5 mindset.

Example CISSP Question

An organization deploys an AI assistant that retrieves information from several internal repositories. The AI service account has access to all documents, although individual employees have different document permissions.

What is the MOST important security requirement?

A. Require employees to use longer passwordsB. Encrypt the AI model's local storageC. Ensure retrieved information is restricted according to the requesting user's authorizationD. Increase the AI model's accuracy

Best answer: C

The central risk is that the AI system could bypass existing authorization boundaries and disclose information users are not permitted to access.

Encryption may protect stored data, but it does not prevent an authorized AI service from returning information to an unauthorized user.

Exam Thinking

The AI should not determine that:

“I can access it, therefore the user can see it.”

The correct principle is:

High-Yield AI Security Rules for CISSP Domain 5

Remember these principles:

1. AI agents and workloads require managed identities.

2. Apply least privilege to every AI identity.

3. Authentication does not imply authorization.

4. AI must not bypass the user's existing authorization boundaries.

5. RAG retrieval must enforce access control before sensitive information is disclosed.

6. Protect AI API keys, tokens, service accounts, and secrets.

7. Avoid shared identities and embedded credentials.

8. Apply stronger controls to privileged AI agents.

9. Use human approval where autonomous actions could create unacceptable impact.

10. Separation of duties still applies to AI-driven workflows.

11. AI-generated identity risk decisions still require appropriate assurance.

12. Deepfakes make strong identity verification increasingly important.

13. Review AI permissions throughout the identity lifecycle.

14. Revoke access when the business requirement ends.

15. Maintain accountability for both human and AI actions.

Final CISSP Takeaway

Domain 5 exposes one of the most important questions in enterprise AI security:

At that point, identity becomes critical.

The CISSP candidate should think beyond whether the AI is authenticated.

Ask:

What identity is it using? What can it access? Whose permissions should apply? What privileges does it really need? Who approves sensitive actions? Can every action be traced?

AI may make access decisions faster and automate more activities, but the fundamental IAM rule remains unchanged:

Trust an identity only to the extent required—and authorize only what the business purpose demands.