DAC, MAC, RBAC, ABAC: How the CISSP Picks an Access Control Model, and the PDP/PEP Machinery That Runs It

Chris Rees
25+ years in IT · Pluralsight author, 4.6/5 across 2,000+ ratings

An auditor asks the exam's model organization a simple question: who can approve a customs declaration over one million euros? The application has 214 permission combinations across 1,400 users, assigned one at a time over nine years, and the answer takes a two-day database query nobody can validate. That is what choosing an authorization model decides: how that question gets answered. Objective 5.4 gives you six models, and the exam tests them by describing how a decision is made and asking you to name the model that makes it that way.
The question each model answers
Every access control model is an answer to three questions: who decides, on what basis, and how expressive the decision can be. Authentication, covered in the Kerberos and RADIUS post, establishes who the subject is. Authorization decides what that subject may do to which object, and the model is the rule for making that decision. The exam's stems almost never name the model; they describe the decision, and the discriminators are the ones in the grid below: an owner granting access is DAC, a label being compared to a clearance is MAC, a job function is RBAC, a global condition is rule-based, a combination of properties is ABAC, and a computed score is risk-based.
DAC: the owner decides
In discretionary access control the object's owner grants access at their discretion, typically through an access control list attached to the object. It is the model of file systems and collaboration tools, and it is flexible precisely because every owner grants independently. That independence is also its weakness: permissions sprawl, nobody can answer the auditor's question, and a Trojan horse running with a user's rights can grant away anything that user owns. The word "owner" in a stem, or a user sharing their own file, points to DAC.
MAC: the system decides from labels
In mandatory access control the system enforces access by comparing the subject's clearance to the object's label, and no user can override it, not even for an object they created. Labels form a lattice: levels such as unclassified through top secret, combined with compartments that implement need to know, so a top secret clearance still does not reach a compartment the subject was not read into. This is the model of government classification and of SELinux-style enforcement, and it is the model the Bell-LaPadula and Biba rules were written to describe. Rigid by design, heavy to administer, and the right answer only when classification is mandated and the cost of a wrong grant is unacceptable.
RBAC: the role decides
In role-based access control, permissions attach to roles and users are assigned roles. It is the commercial default because it scales administration and makes entitlements reviewable: you audit a role once rather than a thousand users. The exam expects the discipline that makes it work.
| Element | Requirement |
|---|---|
| Role definition | derived from job function, never by copying what an individual currently has, which imports every existing excess |
| Role granularity | too few roles forces over-granting; too many recreates per-user administration through role explosion |
| Role hierarchy | senior roles inherit junior permissions, so inheritance is explicit rather than duplicated |
| Static separation of duties | conflicting roles cannot both be assigned to one user; enforced at assignment |
| Dynamic separation of duties | conflicting roles may both be held but not activated in the same session or transaction |
| Role ownership | every role has a business owner who periodically confirms its contents |
Static and dynamic separation of duties are a favorite pair. Static means the person who creates vendors can never also hold the role that approves payments. Dynamic means they may hold both, but cannot use both in the same transaction. A stem about a user who was legitimately assigned two roles but is blocked from exercising them together is dynamic.
Rule-based, ABAC and risk-based: context decides
Rule-based access control applies global rules to every subject regardless of identity: a firewall ruleset, a time-of-day restriction, a source-network condition. It is usually layered on another model rather than standing alone, and the tell in a stem is a condition that applies to everyone.
Attribute-based access control evaluates attributes of the subject, the object, the action and the environment against a policy: a broker may view a declaration if their organization is the declarant, the declaration is not archived, and the request comes from a managed device. It is the most expressive model and the hardest to govern, because a policy that can say anything is a policy nobody can fully review. The exam's most common confusion is RBAC against ABAC. RBAC asks what role the user holds and grants that role's fixed permissions. ABAC asks whether the properties of this subject, this object, this action and this moment satisfy the policy. If the decision depends on properties of the specific object or the current context, it is ABAC; if it depends only on the user's job function, it is RBAC.
Risk-based or adaptive access adds a computed score from signals such as device posture, location, behavior, time and resource sensitivity, and can permit, deny or demand step-up authentication. It is the model behind conditional access and Zero Trust, and it is the answer when a stem describes a decision that changes with circumstances a static rule could not anticipate.
The architecture has four named parts, and the exam tests the split between the first two. The policy enforcement point intercepts the request where the resource is served and enforces whatever answer comes back. The policy decision point evaluates the policy against the supplied attributes and returns permit or deny. The policy information point supplies the attributes: directory data, device posture, data classification, a risk score. The policy administration point is where policy is authored and versioned. XACML is the standard language and request-response model for this architecture. A distributed PEP with a central PDP is exactly what makes Zero Trust implementable, and a stem that asks which component "intercepts" wants the PEP, while one that asks which component "evaluates" wants the PDP.
Matrix, ACLs and capability tables
The access control matrix is the conceptual table of subjects against objects, with each cell holding the rights. Nobody stores the whole table; systems store it by column or by row. Read a column and you have an access control list: attached to the object, listing who may do what to it, which is how file systems work. Read a row and you have a capability list: carried by the subject, listing what it may do to which objects, which is how tokens and tickets work. The discriminator is whether the list is attached to the object or carried by the subject. A Kerberos ticket is a capability; a folder's permissions are an ACL.
Picking the model in a scenario
The exam's selection questions describe an organization and ask which model fits. The answer follows from what the organization needs the decision to depend on.
| Scenario | Model | Why |
|---|---|---|
| Military or intelligence system with classified data | MAC | classification is mandated, users must not be able to share below their level, and the cost of a wrong grant is unacceptable |
| Hospital with clinicians, nurses, billing and researchers | RBAC with an ABAC layer | job function decides most access; record-level scoping (this patient, this ward, under treatment) needs attributes |
| Startup with twenty people and shared documents | DAC | ownership is meaningful, flexibility matters more than auditability, and the administration of MAC or ABAC is unaffordable |
| Cloud API serving many tenants | ABAC | the decision depends on the tenant, the resource's tags, the action and the calling context, none of which a static role captures |
| Any of the above, accessed remotely | risk-based on top | device posture, location and behavior change per request, and step-up authentication answers the change |
The model organization from the opening rebuilt its customs application on RBAC with an ABAC layer for record-level scoping, and the auditor's question became a two-line policy anyone could read. That combination, roles for the coarse grant and attributes for the fine one, is the pattern the exam expects for most modern enterprises, with MAC reserved for mandated classification and DAC for the small and informal.
Four stems, decoded
- A user creates a document and finds she cannot share it with a colleague whose clearance is lower. MAC. The system enforces the label, and ownership confers no right to override it.
- A finance clerk holds both the vendor-creation and payment-approval roles but the system refuses to let him use both within one transaction. Dynamic separation of duties in RBAC. Static would have prevented the assignment.
- A broker can view a declaration only if her firm is the declarant, the record is not archived, and she is on a managed device. ABAC. The decision depends on object and environment attributes, not on a role.
- A login from an unfamiliar country to a sensitive application triggers a request for a second factor. Risk-based access. The score changed the decision; the policy did not.
Key takeaways
- Name what the decision is based on and the model names itself. Owner: DAC. Label against clearance: MAC. Job function: RBAC. Global condition: rule-based. Attributes: ABAC. Score: risk-based.
- RBAC asks what role you hold; ABAC asks whether this subject, object, action and context satisfy the policy. Object or context in the decision means ABAC.
- Static separation of duties blocks the assignment; dynamic blocks the activation. Roles come from job functions, never from copying an individual's access.
- The PEP intercepts and enforces; the PDP evaluates; the PIP supplies attributes; the PAP manages policy. Distributed PEPs with a central PDP are how Zero Trust runs.
- ACLs are the matrix by column, on the object; capabilities are the matrix by row, carried by the subject. Default deny, and watch the broad allow that matches first.
Authorization mechanisms are objective 5.4 of CISSP Domain 5, Identity and Access Management, 13% of the exam, and the bridge between the authentication systems of 5.6 and the IAM fundamentals every later domain assumes. Work all six Domain 5 objectives with our free CISSP Domain 5 study guide.
#CISSP #ISC2 #AccessControl #RBAC #ABAC #MandatoryAccessControl #ZeroTrust #IAM #InfoSec #CyberSecurity
Go past the acronyms
Authentication systems and authorization mechanisms are two of six objectives in CISSP Domain 5, 13% of the exam. Work all 26 topics, from controlling access to assets through federation, the provisioning lifecycle and Kerberos, RADIUS and TACACS+, with 60 practice questions.
- Find your gapsA free readiness quiz scores you domain by domain.
- Study what's on the testWork through every objective, with practice questions.
- Book it with confidenceSchedule the exam when you're scoring well in every domain.
Read a free excerpt from CISSP Domain 5: Identity and Access Management (IAM)
Keep reading
Injection, Broken Access Control, Deserialization, TOCTOU: The Four Flaw Families and the Coding Rules That Kill Them
CISSP Domain 8 does not ask you to write code. It asks you to recognize the mistake in a paragraph and name the rule that prevents it. The four families behind most application breaches, why memory safety is a language choice rather than a coding style, what a race condition really is, the maturity models that measure whether an organization can keep producing secure software, and the metric that tells you whether shifting left worked.
Read CISSP deep divesLife Safety First: Mantraps, Anti-Passback, Duress Codes, and the Physical Security Layer CISSP Won't Let You Skip
CISSP Domain 7 ends with two objectives most candidates skim: managing physical security and personnel safety. They are the source of the exam's most reliable questions. Layered perimeter defense from standoff to badge zones, why a mantrap and anti-passback exist, guards against sensors against barriers, the life-safety rule that overrides every other control, duress codes, travel risk, and the insider and social-media exposures the outline names.
Read CISSP deep divesAssessment vs. Audit vs. Pen Test: Test Types, Team Colors, and Why Independence Decides the Answer
CISSP Domain 6 is small, and its questions hinge on distinctions candidates think they already know. Vulnerability assessment against penetration test, black, white and gray box, what red, blue, purple and white teams actually do, KPIs against KRIs, SOC 2 Type I against Type II, and the single principle, independence, that decides who may perform which test and report to whom.
ReadFollow Skillthropic
Keep learning between study sessions
The same exam-focused teaching, in smaller pieces, wherever you scroll.
Enjoyed this? Get the AI security news that matters.
Join The AI Security Brief for the top AI security news, plus what's important to the C-suite. Free, straight to your inbox.
No spam. Unsubscribe anytime.