
Need to comply with Executive Order 14028 on zero trust? This guide explains how to implement zero trust architecture for stronger security.
What a policy enforcement point (PEP) is, how it works with the PDP, PIP and PAP, where it sits in NIST zero trust architecture, and common PEP types.


Content writer, trio.so
A policy enforcement point (PEP) is the component that sits between a user or device and a protected resource and carries out the access decision: it lets the request through, blocks it, or ends a session that is already open. It does not make the decision itself. It sends the request to a policy decision point (PDP), which checks it against the organization's policies, and then enforces the answer. In NIST's zero trust architecture (SP 800-207), the PEP is the gate in the data path, while the policy engine and policy administrator behind it make up the PDP. In practice a PEP is an identity-aware proxy, an API gateway, a network gateway or an agent on the device.
If you met the term in a zero trust design, an exam or an audit and want to see how the pieces fit together, the components table and the request flow below show it in one pass.
The two names describe a split of duties that goes back to XACML, the OASIS standard for access control policies. One component decides, the other acts. Keeping them apart means the rules live in one place and can be enforced in many.
The PDP evaluates each access request against the policies that apply to it and returns a decision, usually permit or deny, sometimes with conditions. NIST SP 800-162 describes it as the component that "computes access decisions by evaluating the applicable digital policies". To decide, it uses facts about the request: who is asking, from which device, for which resource, and in what context.
The PEP intercepts the request, asks the PDP, and enforces the result. NIST's glossary puts it plainly: it "enforces policy decisions in response to a request from a subject requesting access to a protected object; the access control decisions are made by the policy decision point." If the decision is deny, the request never reaches the resource. If it is permit, the PEP opens the connection and, in a zero trust design, keeps watching it so it can be cut off later.
For example, when an employee opens an internal finance app, the request goes to a gateway in front of the app (the PEP). The gateway asks the PDP, which checks the user's role, the device's state and the time of day, and answers. The gateway then lets the session through or returns an access denied page.
Two more components complete the model, and most PEP and PDP explanations refer to them:
| Component | What it does | Example |
|---|---|---|
| Policy enforcement point (PEP) | Intercepts the request and enforces the decision | An identity-aware proxy in front of a web app |
| Policy decision point (PDP) | Evaluates the request against policy and returns permit or deny | The policy engine in an identity provider or a zero trust service |
| Policy information point (PIP) | Supplies the facts the PDP needs | The user directory, an HR system, device management or threat intelligence |
| Policy administration point (PAP) | Where policies are written and managed | The admin console where security teams define access rules |
In a zero trust architecture, no user or device is trusted because of where it sits on the network. Every request to a resource is checked, and the PEP is where that check is enforced. NIST SP 800-207 describes three logical parts:
Together, the policy engine and the policy administrator form the PDP. Everything on the resource side of the PEP is the implicit trust zone; everything before it is not trusted until the PDP says so. NIST also notes that the PEP can be one component or two: an agent on the user's device and a gateway in front of the resource, working together.
Zero trust decisions are made per request, not once at sign-in. The PDP can take into account the user's identity, the device's health (for example, whether it is managed, encrypted and patched), the location and the sensitivity of the resource. If any of those change during a session, such as a device falling out of compliance, the PDP can revoke access and the PEP ends the connection.
Many PEP and PDP setups use attribute-based access control: policies are written as conditions on attributes of the user, the resource and the environment rather than as fixed lists of who can access what. A rule such as "finance staff can open payroll records from company-managed devices during working hours" is an ABAC policy. The PEP enforces it the same way for every request, and the PIP supplies the attributes.
For instance, a financial firm might let employees open customer account data only from company devices that meet its security baseline. The same employee on a personal laptop is denied, or gets read-only access, depending on the policy.
Whatever form it takes, a PEP needs a few capabilities to do its job well.
A PEP intercepts every request to the resources it protects, asks the PDP and applies the answer, including conditions such as step-up authentication or read-only access. It should fail closed: if it cannot reach the PDP, it denies access rather than letting requests through. It should also log each decision, so security teams can see who was allowed in and who was refused.
Because the decision logic lives in the PDP, PEPs can be placed wherever resources are: in front of web apps, at API gateways, at network edges, inside cloud services and on endpoints. Open source tools are common here, for example API gateways and service meshes that can call an external authorization service, alongside commercial zero trust network access (ZTNA) products.
A hospital runs a cloud-based patient records system. A doctor opens it from a clinic workstation. The request reaches an identity-aware proxy in front of the system (the PEP), which asks the PDP. The PDP checks the doctor's identity and role, whether the workstation is a managed and compliant device, and whether the request comes from an expected network. If all conditions pass, the proxy lets the session through; if the device fails a check, access is denied and the attempt is logged.
PEPs come in several forms, and a zero trust design usually uses more than one:
So what is the best policy enforcement point for zero trust? There is no single answer: the best PEP is the one closest to the resource it protects that can see every request to it. For web apps that is usually an identity-aware proxy; for APIs, an API gateway; for private network services, a ZTNA connector; and for data on the device, an endpoint agent. Many organizations use all four, with one central PDP so the policies stay consistent.
PEPs turn access policy from a document into something that happens on every request. They help in three ways.
Because every request passes through a PEP, a stolen password alone is not enough when the policy also requires a known, healthy device or MFA. Placing PEPs close to each resource also limits how far an attacker can move after getting in. For the common ways attackers get in, see unauthorized access to computer systems.
Frameworks such as HIPAA, PCI DSS and ISO/IEC 27001 expect access to sensitive data to match each person's role and to be logged. PEPs enforce those rules the same way every time and record each decision, which gives auditors evidence that access control works in practice. They support a compliance program; they do not make an organization compliant on their own.
Zero trust asks you to verify explicitly, use least privilege and assume breach. PEPs are where all three meet the traffic: each request is verified, access is limited to what the policy allows, and sessions can be cut off as soon as something changes. See the least privilege access model, network segmentation and defense in depth for the controls around it, and zero trust architecture for the full model.
A policy enforcement point is the gate that carries out access decisions: it intercepts each request, asks the policy decision point, and enforces the answer. In NIST's zero trust architecture it sits in the data path, with the policy engine and policy administrator behind it, and it can take the form of a proxy, a gateway, a connector or an agent. Put PEPs as close to your resources as you can, keep the decisions in one PDP, and feed it good information about users and devices.
trio.so brings device management, patching and security policies for Windows, macOS, Linux, iOS/iPadOS and Android into one console. Enroll a device and see it in the console within minutes.
trio.so brings device management, patching and security policies for Windows, macOS, Linux, iOS/iPadOS and Android into one console. Enroll a device and see it in the console within minutes.





Have questions? We've got answers. This section covers some of the most commonly asked questions related to this topic.
Related
The related industry news, interviews, technologies, and resources.

Need to comply with Executive Order 14028 on zero trust? This guide explains how to implement zero trust architecture for stronger security.

Explore the essentials of a Policy Administration Point (PAP) in cybersecurity. Learn about PAP tools, examples, and how they enhance access control.

Explore policy decision point architecture, meaning, & examples. Learn how PDPs strengthen access control and how MDM optimizes policy management.

Discover the essential differences between ZTNA vs SASE vs CASB. Learn how each solution enhances network security, cloud access, and data protection.

What the Kipling Method is, how zero trust uses its six questions (who, what, when, where, why, how) to write access policy, with a worked example.

Discover the key differences between Zero Trust vs. VPN, their pros and cons, and how you can choose the right security model for your business.

This blog outlines ten best practices to safeguard sensitive information and company resources in the era of remote work.

Explore how Zero Trust Architecture revolutionizes network security by adopting a 'never trust, always verify' approach.
14-day free trial. No credit card required.
trio.so for MSPs: one console, separate clients