Zero trust network access in plain words
Zero trust network access (ZTNA) gives a user access to one resource at a time, after a check of the user and the device. This guide explains the idea, the NIST terms, and how ZTNA differs from a VPN and from SASE.
Each request gets a decision
The Client asks the policy engine for each Resource. The engine knows the user, the Groups, and the device.
No Policy means no path
The engine drops a request that no Policy permits. The Client never learns how to connect.
A permitted request gets keys
For a permitted request, the engine sends keys and connection data to the Client and to a Gateway.
The Gateway enforces
Traffic goes through an encrypted tunnel to the Gateway, and only to the Resource that the Policy names.
What zero trust network access is
Zero trust network access is a way to give people access to private applications. It rests on three ideas.
- No trust from network location. A device does not get access because it is in the office or on a VPN. A request from inside the network gets the same checks as a request from a hotel.
- Verify identity and device on each request. The system checks who the user is, and frequently the state of the device, each time the user asks for a resource.
- Least privilege for each resource. A user gets access to one application or one service, not to a network. Access that nobody granted does not exist.
The older model is a perimeter. A firewall separates a trusted inside from an untrusted outside, and a VPN moves remote users to the inside. That model fails when one account or one laptop on the inside is compromised.
The term ZTNA is used loosely in marketing. Many products with the label are VPNs with a new sign-in page. Judge a product by what it does, not by the label.
The NIST terms in plain words
NIST SP 800-207 is the reference document for zero trust architecture. It names two parts that each design has.
Policy decision point
The policy decision point decides. It receives a request, looks at the identity of the user, the device, and the rules, and then answers yes or no. NIST divides it into a policy engine, which makes the decision, and a policy administrator, which sets up or shuts down the path.
Policy enforcement point
The policy enforcement point acts on the decision. It sits in the path of the traffic. It opens the connection between the user and the resource, monitors it, and closes it.
Why the split matters
The decision is in the control plane, and the traffic is in the data plane. One place holds the rules, so they are consistent. Many enforcement points can sit close to the resources, so traffic does not make a detour.
When you evaluate a product, ask two questions. Find out where the decision is made, and find out where your traffic goes.
Agent-based and agentless ZTNA
Agent-based ZTNA
The user installs a client on the device. The client authenticates the user, reports on the device, and builds an encrypted tunnel to an enforcement point near the resource.
This type works for any protocol, such as SSH, databases, and remote desktop. It can also check the device. The cost is that you must install software on each device.
Agentless ZTNA
The user opens a browser and signs in to a reverse proxy. The proxy checks each HTTP request and sends it to the private application.
This type needs no software on the device, which helps with contractors and unmanaged devices. It works for web applications, and some products add SSH or remote desktop in the browser. The provider usually decrypts the traffic to inspect it, and the device gets fewer checks.
Many organizations use the two types. Employees get an agent, and third parties get a browser.
ZTNA vs VPN, and ZTNA vs SASE
ZTNA vs VPN
A traditional VPN authenticates a user one time and then gives the device a route to a network. A firewall behind the VPN can limit that route, but many setups leave it wide. The VPN concentrator also needs an open port on the internet, which attackers scan for.
ZTNA turns this around. The default is no access. Each resource needs a rule, and the system checks the rule for each connection. A stolen account reaches the resources of one user, not a subnet.
The two can use the same tunnel technology. The difference is the access model, not the encryption.
ZTNA vs SASE
SASE is a bundle of network and security services that one cloud provider delivers. It usually contains ZTNA, a secure web gateway, a cloud access security broker, a firewall service, and SD-WAN.
ZTNA is thus one component of SASE. You can buy ZTNA alone if your need is access to private applications. You need the rest of SASE only if you also want to filter web traffic or to connect branch offices.
Where ZTNA projects go wrong
- One wide rule. A team moves to ZTNA and then adds one rule that gives all users the full network. That setup is a VPN with a different name.
- A big migration. A plan to move all applications on one date usually stops. Move one application at a time, and keep the VPN until the last one moves.
- Identity that is out of date. ZTNA is only as good as its group data. If a person who left stays in a group, the rules still permit them.
- A proxy in the middle. A cloud proxy adds a hop to each connection, and the provider can see the traffic. Know this before you send database traffic through it.
- No device check. A correct password from a compromised laptop is still a risk. Add device conditions for sensitive resources.
How Firezone does zero trust network access
Firezone is an open source, agent-based ZTNA platform built on WireGuard. Its architecture draws on the NIST guidelines, and its parts map to the NIST terms.
- Decision. The policy engine is in the control plane that Firezone runs. It evaluates a Policy each time a Client tries to connect to a Resource, and again when the Policy changes.
- Enforcement. Clients run on your devices, and Gateways run in your networks. A Gateway carries traffic only for connections that the policy engine approved.
- Default-deny. A new account has no Policies, so nobody can reach anything. Each Policy gives one Group access to one Resource.
- Your identity provider. Users sign in through OIDC with the provider that you already use. Directory sync keeps users and Groups current.
- Direct encrypted paths. Traffic goes from the Client to the Gateway in a WireGuard tunnel. If no direct path exists, a relay forwards the encrypted packets. Firezone cannot decrypt them.
If the policy engine refuses a request, it drops the request. The Client does not learn how to connect to the Gateway, and no tunnel exists.
Firezone works at layer 3, so it carries any IP traffic. It runs beside an existing VPN. You can start with one CIDR Resource that mirrors the VPN, and then move one application at a time to narrow Resources.
Firezone does not offer clientless browser access, and it is not a SASE suite. It does one job, which is access to private resources.
A legacy VPN, cloud-proxy ZTNA, and Firezone
The three approaches differ in the access model and in the path that your traffic takes.
| Legacy VPN | Cloud-proxy ZTNA | Firezone | |
|---|---|---|---|
| Default-deny for each resource | NoAccess to a network | Yes | YesOne Policy for each Group and Resource |
| Where the decision is made | On the VPN concentrator, at sign-in | In the cloud of the provider | In the Firezone policy engine, for each connection |
| Where your traffic goes | Through the concentrator | Through the cloud of the provider | Direct to your Gateway, or through a relay |
| The provider cannot read your traffic | YesYou run it | NoThe proxy usually decrypts it | YesEnd-to-end WireGuard encryption |
| No inbound port on your firewall | No | Yes | Yes |
| Any protocol, not only web | Yes | PartialBest for HTTP | YesTCP, UDP, and ICMP |
| Access with no client on the device | No | YesThrough a browser | NoA Client is necessary |
The table describes the usual design of each category. A specific product can be different. Read the documentation of the product before you decide.
Common questions
What is zero trust network access (ZTNA)?
ZTNA is a way to give users access to private applications one resource at a time. The system checks the identity of the user, and frequently the device, for each request. Network location gives no access by itself.
Is ZTNA a replacement for a VPN?
For remote access to private applications, yes. ZTNA does the same job with narrower access and no open inbound port. Most teams run the two side by side for some time and move one application at a time.
What is the difference between a policy decision point and a policy enforcement point?
The policy decision point decides if a request is permitted. The policy enforcement point sits in the path of the traffic and applies that decision. NIST SP 800-207 defines the two terms.
Is ZTNA the same as SASE?
No. SASE is a bundle of cloud services that includes ZTNA, a secure web gateway, and SD-WAN. ZTNA is the component that controls access to private applications. You can use ZTNA with no other SASE component.
Does ZTNA need an agent on each device?
Not always. Agentless ZTNA uses a browser and a reverse proxy, and it mostly serves web applications. Agent-based ZTNA needs a client, and it serves any protocol and can check the device. Firezone is agent-based.