VPN vs ZTNA: remote access for organizations
A VPN connects a device or a whole site to a network, and what the user can reach afterwards depends on firewall rules behind the tunnel. ZTNA (zero trust network access) connects an identified user on a checked device to one application at a time, and evaluates each session against policy. Most organizations end up running both: site-to-site VPN between locations, and ZTNA or a tightly scoped VPN with MFA for people and vendors.

Two questions that get mixed together
“VPN or ZTNA” is usually asked as one question. It is two.
The first is how locations connect to each other: branches, plants, data centers and cloud networks. The second is how people and outside vendors reach internal systems from wherever they are. A site-to-site VPN answers the first. A remote access VPN or ZTNA answers the second. Choosing ZTNA for staff does not remove the tunnels between your sites.
The two kinds of VPN
Site-to-site VPN
A site-to-site VPN is an encrypted tunnel between two gateways, typically firewalls or routers. It is always on. Devices at each site do not know it exists. They send traffic to an address at the other site and the gateways handle encryption and routing.
Typical uses are branch to headquarters, office to a cloud virtual network, and office to a hosting provider. The design work is in the routing, the address plan (overlapping subnets between sites are a common obstacle), failover between internet lines, and the firewall rules on each end. A tunnel that permits everything between two sites joins them into one flat network, which means a problem at one site can spread to the other.
Remote access VPN
A remote access VPN is started from a user’s device with a client application. The user authenticates, the device receives an internal address, and traffic to internal systems goes through the tunnel.
What the user can reach after connecting is decided by firewall rules behind the VPN gateway. In many installations those rules are broad because they were written once and never narrowed. That is the real weakness of the model. The tunnel itself is sound. The issue is that a connected laptop sits on the network, and if the laptop or the account is compromised, the attacker sits there too.
The VPN gateway is also exposed to the internet by design. It must be patched quickly when the vendor publishes a fix.
What ZTNA changes
ZTNA is an industry term for products that apply zero trust principles to remote access. The reference document for zero trust is NIST Special Publication 800-207, published in August 2020. It describes zero trust as assuming “no implicit trust granted to assets or user accounts based” on their network location, and lists tenets that include “all communication is secured regardless of network location” and “access to individual enterprise resources is granted on a per-session basis”.
In a ZTNA product those principles turn into three practical differences from a VPN.
Access is per application. The user is connected to a named application, such as the ERP web interface or one server’s remote desktop. They are not given an address on the internal network. Systems outside their policy are not reachable and usually not visible.
The decision uses identity and device state. Each session is checked against policy: who the user is, which group they belong to, whether the device is managed, encrypted and patched, and sometimes where and when the request comes from. The check is repeated, so a device that falls out of compliance loses access.
Applications are not exposed to the internet. In most designs a small connector inside your network makes an outbound connection to the access broker. There is no inbound port to scan. The broker itself is the exposed part, and it is run either by the ZTNA vendor or by you.
NIST describes zero trust as a set of principles and not a single product, so two ZTNA products can work quite differently. Some route traffic through the vendor’s cloud. Others are self-hosted. That choice matters for latency, for data residency and for what happens when the vendor has an outage.
VPN and ZTNA side by side
| Question | Remote access VPN | ZTNA |
|---|---|---|
| What the user connects to | The network | One application per policy |
| What limits reach after login | Firewall rules behind the gateway | The access policy itself |
| Device health checked | Sometimes, at connect time | Typically at connect time and during the session |
| Exposure to the internet | VPN gateway accepts inbound connections | Usually outbound-only connectors |
| Works with any protocol | Yes, it is a network connection | Depends on product, strongest for web, SSH and remote desktop |
| Connecting whole sites | Yes, with site-to-site tunnels | No |
| Setup effort | Lower at the start | Higher, each application must be defined |
| Dependence on a third party | Low when self-hosted | High when the broker is a cloud service |
The setup effort row deserves attention. ZTNA requires an inventory of applications and who needs each one. Organizations that have that inventory deploy quickly. Organizations that do not have it discover the gaps during the project. The inventory is worth producing either way, because the same list tightens VPN firewall rules.
MFA comes first in both models
Multi-factor authentication is the highest-value control for remote access, whichever model you run. A remote access system that accepts a password alone is one phishing email away from an outsider on the inside.
Practical points:
- Enforce MFA at the identity provider and have the VPN or ZTNA product authenticate against it, so there is one place to manage accounts and one place to disable a leaver.
- Prefer phishing-resistant methods such as hardware security keys or passkeys for administrators. App-based codes are a reasonable baseline for everyone else.
- Remove local accounts on the VPN gateway, apart from one documented emergency account with a long random password stored offline.
- Service accounts that cannot use MFA should not be able to log in remotely at all.
MFA and access policy are part of a wider program, described on our cybersecurity services page.
Vendor and contractor access
Outside vendors are the hardest remote access case. They support a specific machine, controller or application. They use their own laptops, which you do not manage. Several of their staff may share one login.
The controls that work:
- Named accounts per person, with MFA. No shared logins.
- Access to the specific systems under the support contract and nothing else.
- Time limits: access enabled for a maintenance window or on request, then closed.
- Session logging, and session recording for production and financial systems.
- A path that does not require the vendor’s unmanaged laptop to join your network, for example a browser-based session through a ZTNA portal or a hardened jump host.
ZTNA handles this case well because access is already per application and per person. A VPN can do it too, with a dedicated vendor group, narrow firewall rules and a jump host, but it takes discipline to keep the rules narrow over the years.
When a VPN is still the right choice
- Connecting sites. Branch, plant, data center and cloud connectivity is a site-to-site problem. ZTNA does not address it.
- Legacy and industrial systems. Applications that use unusual protocols, broadcast traffic, or connections initiated from the server to the client often do not work through a ZTNA broker.
- Administrators who need broad network access. Network engineers troubleshooting routing need a real network connection. Give it to a small group, with MFA and logging.
- No tolerance for a cloud dependency. If remote access must work when a vendor’s cloud is down, or data may not pass through a third party, a self-hosted VPN or a self-hosted ZTNA product is the option.
- A small number of remote users and applications. A patched VPN with MFA and per-group firewall rules can be the proportionate answer.
A staged way to move
A full replacement in one step is rarely necessary. A sequence that lowers risk at each stage:
- Turn on MFA for every remote access path that lacks it.
- Inventory who connects, from which devices, to which systems. Remove accounts nobody can explain.
- Narrow VPN firewall rules by group, starting with vendors.
- Move vendors and contractors to per-application access.
- Move staff applications to ZTNA in batches, web applications first.
- Keep a small VPN for administrators and the systems that cannot move. Keep site-to-site tunnels, with rules between sites reviewed.
Steps 1 to 3 need no new product. They are also where most of the risk reduction is.
Who runs it afterwards
Remote access is an operational system. Certificates expire, gateways need firmware, policies drift as people change roles, and vendors come and go. Someone has to own the access list and review it on a schedule. In most organizations that sits with the team that runs the firewalls and switches, which is why we treat it as part of network management services and of a wider managed IT services retainer.
What to do next
Export the list of accounts that can connect remotely today, mark which have MFA, and mark which belong to vendors. That list shows whether your first project is MFA, vendor access or a redesign. Our VPN and remote access services cover site-to-site design, remote access VPN and ZTNA implementation. Goldenroz is independent and holds no partner status with any firewall or ZTNA vendor. Projects start at $25,000 at a fixed price per phase, and ongoing management is a monthly retainer quoted after an assessment.