When a customer says they want to replace their VPN with zero trust network access (ZTNA), the conversation usually jumps to features, pricing and which products to compare. What rarely comes up is what kind of architecture they are signing themselves up for. 

That is a bit like choosing a restaurant without asking whether you are dining in or ordering delivery. The food might be the same, but the experience, the risks and what you can actually get are not.

In ZTNA, the two options are network-based and proxy-based. Each carries its own type of risk, and each suits different apps and different organizations. 

Today most products offer both, but that is not a reason to skip the question. 

Hybrid ZTNA is not one thing. Where a product started shapes where it is strongest, and Cisco took an entirely different route. 

This is a guide to how the two models work, the kinds of risk that come with each, and how to figure out what you actually need.

Quick reset: ZTNA is not zero trust

This one trips people up. Zero trust is the strategy, the why. Segmentation is the architecture, the how. ZTNA is the enforcement, the point where those principles meet a real user trying to open a real app. It completes the model. It does not replace it.

And why do we need it? The legacy VPN. A VPN checks who you are once, then hands you network access, so the app you wanted becomes one of many things you can reach. Add a flat network, group-based access and trust that never gets re-checked and one stolen credential turns into a long walk through your environment. Vendors and partners often get the same access as employees, which makes it worse.

Two ways to get a user to an app

Both models try to do the same thing: get a user to the app they are approved for, and nothing else. The difference is where the control point sits.

Proxy-based ZTNA: Ordering delivery

Diagram of proxy-based ZTNA as food delivery, where a user reaches a courier acting as the access broker, which connects only to the payroll app while the finance database and HR app have no path.
Proxy-Based ZTNA


With delivery, you never see the restaurant. You never go inside, and you don't even need to know its address. You reach out to a delivery service like DoorDash and place an order, and the driver brings back only what you asked for. You have no access to the kitchen or the bar, because the restaurant never opens its doors to you.

That is how proxy-based ZTNA works. The user and the app each connect outbound to an access broker in the middle, and the broker joins them for that one app. The user never needs to know where the app lives; nothing connects inbound to the data center; and the user has no path to applications they are not approved for. 

Network-based ZTNA: Dining in
 

Diagram of network-based ZTNA as dining in, where a user passes a security gateway host stand to reach the payroll app, while the finance database and HR app are exposed if rules are broad.
Network-Based ZTNA

With dining in, you walk up to the host stand, get checked and come inside. Once you are in the building, the layout and the staff decide where you can go. You are allowed to go to your table, but the kitchen and the bar share the building, so if the rules are broad, you could end up there too.

The upside is that you are really there. A server can walk over to you, and you can make off-menu requests: swap a side, hold the onions, ask the kitchen for something special. Delivery works best when you order from the standard menu.

That is how network-based ZTNA works. The user connects through a tunnel to a security gateway, and firewall policy decides which apps are reachable. Because it supports inbound connections, it handles the legacy, server-initiated and peer-to-peer apps that a broker cannot. They often expect unusual protocols, open ports or the app reaching back to the user, and that is easier when you are in the building.

In the proxy model, the architecture does the work: no path exists unless policy creates one. In the network model, zero trust comes from how you design it, through segmentation, firewall rules and decisions about who can reach what. By default means less to build. By design means more control, and more for your team to own.

Every approach carries its own kind of risk

Neither model is risk-free. They just carry different types of risk. For network-based ZTNA, the inherent risk is reach. For proxy-based ZTNA, it is policy.

Network-based risks: Allowed to one app, reachable to many.

  • Broad allow rules. Policy that grants a subnet or app group, not a single app, so one approval covers many systems.
  • A shared network path. The tunnel gives network-level reach, so anything on that segment can be probed.
  • No east-west inspection. Traffic between servers never crosses the gateway firewall, so pivots go unseen.

The firewall only controls what crosses it. Anything behind the tunnel is one pivot away.

Diagram of network-based ZTNA in which a compromised user passes the security gateway over an IPsec tunnel to the payroll app, then reaches the HR app and finance database through lateral movement.
In network-based ZTNA, one compromised user can spread from an allowed app to others behind the tunnel.

Proxy-based: The paths exist only if policy creates them.

  • Over-broad rules. A rule like "all internal apps" creates a private path to every app it covers.
  • Policy drift. Rules pile up as teams and roles change, and old access rarely gets removed.
  • Unclassified apps. Policy can only protect what it knows about, so apps with no owner tend to inherit a default rule.

The architecture removes the extra paths. Policy decides whether they come back.

Diagram of proxy-based ZTNA in which a broad rule covering all internal apps gives a user unintended access to the HR app and finance database along with the payroll app.
In proxy-based ZTNA, a broad rule can reopen paths to apps the user was never meant to reach.

Hybrid is not one thing

Almost no customer fits cleanly on one side. They have legacy systems that need inbound connectivity, a long tail of web and SaaS apps, and contractors that are far simpler via a proxy. So most products now offer both. But hybrid ZTNA vendors are not all created equal, and for most, their strength comes from their heritage.

Proxy heritage (Zscaler, Netskope)

These started as proxy-based services and added network-style access.

  • The benefit: These vendors excel in web and SaaS-style private apps, and contractors on unmanaged devices.
  • The consideration: Look closely at how they handle legacy, server-initiated and peer-to-peer apps. That is where a broker model has the most to prove.
  • How they frame it: They frame the network-style option is a workaround for certain apps and protocols, not the first choice.

Network heritage (Cato, Fortinet)

These started from the network and added proxy-style access.

  • The benefit: These vendors excel in legacy, server-initiated and peer-to-peer apps, plus site-to-site connectivity and microsegmentation.
  • The consideration: Look closely at the experience for web and SaaS apps and for contractors and at how much configuration your team will own.
  • How they frame it: Network-based ZTNA is just as important and secure as proxy-based ZTNA, a full option and not an add-on.

Cisco took a different route

Instead of growing from one side, Cisco designed both paths into a single service, with the choice per app made at the edge: ZTNA via resource connectors, or VPN as a service over network tunnels.

  • The benefit: an onramp from VPN to ZTNA at your own pace.
  • The consideration: two paths mean two things to operate, so ask how posture checks and policy line up across them.
  • How they frame it: VPN as a service is a way to more easily on-ramp proxy-style ZTNA from VPN, and that transition is the point.

Two more worth knowing

  • Palo Alto Prisma offers network-level ZTNA only. As a heritage firewall company, it views that approach as valid in its own right. The question is whether it fits your apps and your organizational ecosystem.
  • Appgate is a standalone ZTNA vendor, network-based by design, with cloud-hosted, self-hosted or isolated deployment. It suits customers who want control and ownership of their infrastructure, and who pair it with a separate web and SaaS security layer.

The framing matters as much as the feature list. It shows you what each vendor sees as the main event, and that tends to show up in where they put their effort. None of this makes one vendor better than another. It makes some a better match for your apps, your team and your risk tolerance.

Okay, but which one is right for you?

There is no universal winner, and I would be wary of anyone who claims one. Here are a few questions to ask yourself as you approach a VPN replacement initiative:

  • Are your hardest apps legacy, server-initiated or peer-to-peer? 
  • Is your world mostly web and SaaS, with lots of contractors and BYOD? 
  • Where did the vendor start? Heritage tells you where to expect strength, and where to test.
  • How does the vendor frame the model you will rely on most? Ask whether it is a workaround, an equal option or an onramp.
  • Who is running it? A mature network team may want the control. A lean security team will value fewer moving parts.
  • How good is your app inventory? Both models depend on good policy, and a broad rule will undo either architecture.
  • Are you moving off VPN over several years? An onramp that lets you move apps over at your own pace may matter more than a clean end state.

One more tip from the field

ZTNA programs rarely stall for technical reasons. They stall because teams try to do everything at once, defer legacy until it becomes a roadblock, skip discovery or leave ownership with IT alone. Start with one high-value use case, like third-party access. A visible win builds momentum for the next project.

Plenty of restaurants now offer both dine-in and delivery, and the good ones did not get there by arguing about which is better. They got good at running both. ZTNA is heading the same way.

For a more detailed and customized assessment of the SSE vendor landscape, reach out to our team.

Technologies