The VPN has secured enterprise remote access for more than two decades, and the model underneath it has barely changed. The reason enterprises are replacing it has less to do with the technology failing than with what it assumed.

A VPN assumes there is a corporate network worth connecting to, that most applications sit inside it, and that remote access is the exception. Few enterprises still operate that way. Applications are split across data centres, private cloud and SaaS. Contractors and unmanaged devices need individual systems, not a network. The gap between what a user needs and what a VPN session grants has become difficult to justify.

Zero Trust Network Access is the practical response. It changes what access means: not a position on the network, but a brokered connection to a named application, evaluated each time it is requested.

Why VPNs worked for traditional enterprise networks

A VPN creates an encrypted tunnel between a user’s device and the corporate network. After authenticating against a gateway, the device is issued an internal IP address and routed as though it were physically in the office, reaching applications through internal DNS and internal addressing.

How VPN access is granted

• Authentication happens once, at connection time.
• Access is granted at the network layer, not the application layer.
• The gateway must be internet-facing, because that is how remote users reach it.

Why that model fitted the networks of the time

• Staff worked from offices. Remote access was the exception, used by a minority of users on company-issued laptops.
• Applications sat in a controlled data centre. Everything a user needed was behind the same perimeter.
• Network-level access was simpler to run. One tunnel replaced per-application access control across dozens of systems.
• Legacy applications needed no changes. Systems with no modern authentication support worked as they were.

The VPN did not become a problem because it was badly designed. It became a problem because every one of those conditions has since changed.

Why traditional VPNs are no longer enough for modern enterprise networks

Three architectural issues account for most of the pressure enterprises are under. Each is a consequence of the design rather than a defect in any particular product.

VPN access is granted at the network layer, not the application layer

Once a session is established, the device is on the network. Control from that point depends on:

• Internal segmentation between network zones.
• Firewall rules between VLANs.
• Permissions inside each individual application.

In practice, segmentation is coarser than intended and rules accumulate without review. Three consequences follow

• A user who needs one application often has reachability to considerably more.
• Credential compromise yields a network position rather than access to a single system, which is where lateral movement begins.
• Access reviews become difficult, because what a user can reach is defined by network topology rather than by a policy anyone can read.

Internet-facing VPN gateways are a permanent target

A VPN concentrator must accept inbound connections from anywhere, because that is how remote users reach it. That means:

• A public IP address.
• Open listening ports.
• A presence that can be found by scanning.

The breach data reflects it. Verizon’s 2025 Data Breach Investigations Report found vulnerability exploitation behind 20% of breaches, with edge devices and VPNs rising from 3% to 22% of those exploitation targets in a single year, a near eight-fold increase.

The operational burden that it creates:

• The gateway has to be patched quickly, because any weakness in it is a weakness in the front door.
• Its authentication has to withstand credential stuffing and MFA fatigue attempts.

Backhauled VPN traffic degrades cloud and SaaS performance

VPN routing sends user traffic to the gateway first. When the destination is a SaaS platform, traffic travels from the user to the concentrator, out to the internet, and back again. A user may be geographically close to the provider’s nearest point of presence and still be routed through a data centre on another continent.

This carries two costs:

• Performance. Latency and packet loss degrade the applications people use most, particularly voice and video.
• Capacity. Gateway sizing must cover peak concurrent load, driving hardware, licensing and refresh cycles for traffic that gains no security value from the detour.

Split tunnelling reduces the performance penalty but reintroduces an inspection gap, which is a trade rather than a fix.

How ZTNA grants access without exposing the network

The most useful thing to understand about ZTNA is not its policy language. It is the direction the connections travel.
A lightweight connector is deployed alongside the application, in the data centre or cloud environment where it runs. That connector opens an outbound connection to the ZTNA broker and holds it open. The application publishes no public IP address and accepts no inbound connections from the internet, so no ingress rule is required at the perimeter.
The user side also connects outbound to the broker, either through a client or, for web applications, through the browser with no client installed.

The three ZTNA access checks: Identity, device and policy

• Identity. The user is authenticated against the enterprise identity provider, typically through SSO with MFA enforced there.
• Device. The device is checked against posture requirements such as management status, disk encryption, patch level and endpoint agent presence.
• Policy. The broker decides whether this identity, on this device, in this context, is entitled to this specific application. Access is denied by default and granted only where a rule permits it.

Only when all three pass does the broker join the two outbound connections into a session to that one application. Nothing else becomes reachable, because no route to the network was created.

What application-level access changes in practice

• Applications cannot be found by scanning. An application that never accepts inbound connections is not discoverable by an attacker probing the perimeter.
• Trust does not persist for the life of a tunnel. Evaluation happens per session, so a change in device posture or policy affects the next request.
• Exposure is consolidated, not eliminated. The broker and identity provider remain internet-facing by design. What changes is that two components are exposed and patched as a service, rather than every gateway and application an organisation runs.

ZTNA vs VPN: What actually changes in the access model

Both models provide secure remote access. The difference is not a list of features but where access is decided and what a session can reach once it is granted.

Where traditional VPNs are still required

ZTNA replaces VPN for user access to applications. It does not replace every use of VPN.

• Legacy applications: Systems relying on fixed internal addressing, server-initiated connections, or protocols such as SMB and RPC often expect full network reachability. Some cannot be moved without redevelopment.
• Infrastructure administration: Engineers working on switches, firewalls, hypervisors and out-of-band management need a management network, not a single application. Many organisations retain a hardened VPN path for this, restricted to a small group.
• Operational technology: SCADA systems, PLCs and industrial control equipment were built for isolated networks, and vendor support terms often rule out the agents ZTNA requires.
• Site-to-site connectivity: Interconnecting branches and data centres is a different problem from user access. Site-to-site VPN and SD-WAN address it.
• Mergers and transitions. When two directories and application estates must communicate before integration completes, VPN is the pragmatic bridge.

Is ZTNA more secure than VPN?

ZTNA is more secure than VPN for remote user access to applications. The difference is architectural rather than a matter of configuration quality, which is what makes it secure by design.

• The application is not reachable from the internet. With no public listener and no ingress rule, there is nothing for an attacker to scan, fingerprint or exploit at the perimeter. A VPN gateway, by contrast, has to be findable in order to work.
• A compromised session reaches one application, not a network. Access is granted per application, so stolen credentials no longer deliver the network position that lateral movement depends on.
• Trust is re-established continuously. Identity, device posture and policy are evaluated at each request rather than once at login, so a device falling out of compliance affects the next session rather than persisting until the tunnel drops.
• Access is denied by default. Applications a user has no policy for are not restricted, they are invisible.

Together these close the three weaknesses that make network-level access difficult to secure: an exposed gateway, excessive reach after authentication, and trust that never expires.
ZTNA delivers most where identity foundations are strong. Complete MFA coverage and an accurate directory are what turn the architecture into enforced policy, which is why identity readiness is usually the first thing to assess.

Operational impact of replacing VPN with ZTNA

The operational case is separate from the security one, and it is often what carries the decision internally.
Under the VPN model, access is expressed as network reachability. Granting it means editing routing and firewall configuration; revoking it means finding and removing those entries. Every new contractor, partner or acquired business adds to a rule base that grows faster than anyone reviews it.

With ZTNA, access becomes a policy assignment tied to a group in the identity provider. What that changes:

• Onboarding: A new starter receives the entitlements their role requires on day one, rather than an account, client deployment and a set of firewall rules.
• Leavers: Removing someone from the directory removes their access, because entitlement is held against identity rather than in a separate rule base.
• Third-party access: Contractors can be scoped to a single application, frequently through a browser with nothing to install, with an expiry date attached.
• Capacity planning: Concentrator sizing and refresh cycles apply only to the smaller remaining VPN population.
• Support load: A significant share of remote access tickets concern connecting, reconnecting and performance while tunnelled. Those disappear for migrated applications.
• Audit evidence: Records show which user reached which application from which device, which is closer to what auditors ask for than gateway session logs.

How ZTNA improves scalability and access governance

Two of these matter more than the rest at board level.

Scalability stops being a hardware question

Supporting more remote users, opening an office or absorbing an acquisition no longer means sizing and procuring gateway capacity. Adding a user group or a private application becomes a configuration task rather than a procurement cycle.

Access becomes explainable

Under a network-centric model, answering who can reach a given system means reconstructing it from VPN groups, routing, segmentation and firewall rules maintained by different teams over several years. Under identity-based access control, access is a stated relationship between a user, a device condition and an application. For customer security assessments, regulatory review and insurance questionnaires, that clarity is frequently what closes the internal argument.

What to assess before replacing VPN access

Most difficulty in these projects comes from discovering constraints late. Five things are worth establishing before committing to a scope.

Application inventory

What is actually reached over the VPN today, by whom, and how often. Gateway logs usually surface applications nobody had on the list.

Protocol dependencies

Which applications rely on fixed internal addressing, server-initiated connections or legacy protocols. This determines what can move and what stays.

Directory consolidation

Whether there is a single authoritative identity provider or several, and whether group membership reflects current roles. ZTNA policy is only as precise as the directory it reads from.

MFA and SSO coverage

Which applications and user populations are covered today, including contractors and service accounts. Gaps here become policy gaps.

Device posture visibility

Whether endpoint management or EDR can report the signals policy will rely on. Clientless access provides limited posture signals, because endpoint state cannot be inspected without an agent, so policies covering unmanaged devices should be scoped accordingly.

The output is a split: applications that move first, applications that move with work, and applications that stay on VPN. Running both in parallel during the transition is the normal approach rather than the exception.

How ZTNA fits alongside SASE and SSE

ZTNA is one capability within Security Service Edge (SSE), alongside Secure Web Gateway (SWG), Cloud Access Security Broker (CASB) and Firewall-as-a-Service (FWaaS). SSE forms the security side of SASE, which adds SD-WAN for network connectivity.

ZTNA does not require a full SASE programme. It can be deployed on its own to address remote access, and it is a common starting point because the problem is well defined and the scope is contained. Organisations that later adopt further SSE capabilities gain shared policy and identity context across them, but that is a sequencing benefit rather than a prerequisite.

Orixcom’s Zero Trust Network Access applies this model to enterprise remote access, brokering identity- and device-verified connections to individual applications rather than extending the network to every user. It can be deployed as a standalone remote access replacement or as part of a broader SSE or SASE rollout, depending on where the immediate operational pressure sits.

Conclusion

The shift underway is from access defined by network position to access defined by identity, device and policy. For most organisations, ZTNA is now the more appropriate model for connecting distributed users to applications, because it takes the application out of internet reach, limits each session to what policy permits, and re-evaluates rather than assumes.

That shift does not require removing VPN from the estate, and most enterprises will not. It requires being deliberate about what each model is for. Network-level access should be reserved for the cases that genuinely need it, and everything else should move to access granted one application at a time.

FAQs

Why are traditional VPNs being replaced by ZTNA in enterprise networks?

Because the assumptions behind the VPN no longer hold. Applications have moved to cloud and SaaS, remote work is the norm, and the gateway must stay internet-facing to function. ZTNA instead brokers access to one named application at a time, verified against identity and device posture rather than network position.

How does ZTNA reduce lateral movement?

Lateral movement needs a network position, and ZTNA never creates one. A session reaches a single authorised application, so stolen credentials deliver that application and nothing adjacent. Applications with no matching policy are not discoverable.

Why are VPNs less suitable for hybrid workforces?

A VPN assumes remote access is occasional. When most staff are remote, gateway capacity must cover near-total concurrency, SaaS traffic is backhauled for no benefit, and every session adds another device to the internal network.

What is the best alternative to a VPN?

It depends on the use case. ZTNA replaces user access to applications. SD-WAN replaces site-to-site connectivity. Where internet and SaaS traffic also needs inspection, ZTNA is deployed alongside SWG and CASB within SSE.