In most organizations, the remote access decision is reduced to the question “which VPN client should we install?” The real question is different: are you opening the whole network to the remote user, or only the applications they need? That’s where the real difference between VPN and ZTNA (Zero Trust Network Access) lies. Below we cover the difference between the two models, ZTNA’s limits and when a hybrid approach makes sense, with a decision tree.
The difference in access model between VPN and ZTNA
A remote access VPN (SSL VPN or IPsec) gives the authenticated client an IP address on the corporate network and makes the client part of the network. Which systems the user can reach is determined by firewall rules written from the VPN pool to the internal network. These rules are mostly written not per user group but per pool subnet. This is the common picture: an “any” permit from the VPN pool to the server network. As a result, most remote users see a much wider network than they need; a compromised VPN account or a connected laptop infected with malware moves laterally within that breadth.
ZTNA grants access at the application level. The user and device are evaluated by the policy engine on every connection request; if permission is granted, a session is established only to that application. Applications aren’t exposed directly to the internet; usually a connector on the internal network makes an outbound connection, and there’s no need to open inbound ports. NIST SP 800-207 defines this model as making the resource access decision per session based not on network location but on identity, device state and context.
The key difference is this: with VPN, trust is granted at connection time and lasts for the session; with ZTNA, trust is re-evaluated on every application access according to policy conditions. One opens the door. The other opens the room.
Agent-based and agentless ZTNA architectures
ZTNA products offer two basic forms of access; most organizations use both:
- Agent-based (client-initiated): A client installed on the endpoint identifies the user and the device, passes device posture information (disk encryption, EDR status, operating system version, whether it’s a managed device) to the policy engine and tunnels traffic only for permitted applications. It’s required for RDP, SSH, file sharing and non-web protocols.
- Agentless (service-initiated / browser-based): The user accesses a web application with a browser through a reverse proxy. It suits unmanaged devices, vendors and business partners; however, device posture visibility is limited and support for non-web protocols is restricted.
Architecturally, the second distinction is where the policy enforcement point sits: the vendor’s cloud service (SSE/SASE platforms) or a gateway in the organization’s own data center. In the cloud-based model, traffic passes through the vendor’s infrastructure; address this separately in the data residency and privacy assessment.
Comparison table: security, performance, operations
| Criterion | Remote access VPN | ZTNA |
|---|---|---|
| Access scope | Network / subnet | Implementation |
| Trust evaluation | At connection time | On every access, continuously |
| Device posture check | Product-dependent, usually at connection time | Part of the policy |
| Exposed surface | The VPN gateway is exposed to the internet | Applications are hidden; the connector connects outbound |
| Lateral movement risk | High, if rules are broad | Low, limited to the application |
| Protocol support | All IP traffic | Product-dependent; UDP and server-to-client connections may be limited |
| Initial setup effort | Low | Requires an application inventory and policy design |
| Operational load | Depends on the rule set | Adding applications and maintaining policy is ongoing |
Generalizing about performance is misleading. With a full-tunnel VPN, all traffic is carried to the data center; a split tunnel reduces this load but leaves internet egress uncontrolled. In cloud-based ZTNA, latency depends on the location of the vendor’s point of presence closest to the user and the connector on the path to the application. In the pilot, measure from the locations where your own users are. A datasheet isn’t enough.
User experience is also a decision criterion. With VPN, users forget to connect, the connection drops, a hotel network blocks IPsec traffic; most remote access support tickets come from these. Agent-based ZTNA runs continuously in the background, and the user just opens the application. On the other hand, when access is denied, the reason (device non-compliance, missing group membership, a policy condition) needs to be shown clearly to the user and the help desk. If it isn’t, troubleshooting becomes harder than with VPN.
ZTNA’s limitations and scenarios where it doesn’t fit
Assuming ZTNA meets every remote access need is the most common mistake migration projects make. Look especially at these situations:
- Server-to-client initiated connections: Help desk tools connecting to a remote computer, management tools such as SCCM reaching the client, VoIP and some active FTP scenarios may not be supported by ZTNA products or may need additional configuration.
- Legacy applications with unclear dependencies: Client-server applications that use dynamic ports, have IP addresses hard-coded or talk to many back-end systems make application-based definition difficult.
- Site-to-site connectivity: Persistent connectivity between branches and data centers isn’t ZTNA’s concern; it’s solved with site-to-site IPsec or SD-WAN.
- Network administrator access: Network teams’ access to broad management networks doesn’t fit easily into the application-based model; this access is usually managed separately through PAM or a jump host.
- Dependency and cost: Cloud-based ZTNA creates dependency on the vendor’s service continuity; how remote access will be provided during a service outage should be planned in advance.
Decision tree: VPN, ZTNA or hybrid
When the questions below are answered in order, the architecture becomes clear in most organizations:
- Can an inventory of the applications remote users access be built? If not, inventory first. ZTNA policy can’t be written without an inventory. In the meantime, the existing VPN is protected with the hardening below.
- Are most applications web-based or SaaS? If so, ZTNA (including agentless access) is a strong candidate; it delivers a quick win for unmanaged devices and vendors.
- Are there groups that need server-to-client connections, a legacy client-server application or broad administrative access? If so, VPN (or PAM) stays for these groups a while longer: the hybrid model.
- Are devices managed? Agent-based ZTNA and posture checks on managed devices; only agentless, narrow-scope access on unmanaged devices.
- Is there vendor and third-party access? If this access is given through a VPN account, this is the first group to move to ZTNA; application-based, time-limited access directly reduces risk.
In practice, most mid-sized organizations stay in a hybrid model for a while: general users and vendors on ZTNA, IT administrators and specialized application users on a hardened VPN. To maintain the enterprise cybersecurity level during the hybrid period, you need to strengthen the VPN with MFA, device checks and rule separation by user group; details are in the enterprise VPN security controls article.
A phased move from VPN to ZTNA
- Visibility: From VPN logs and firewall session records, work out which user group accesses which applications. This list is the draft of the ZTNA application catalog.
- Preparing the identity and device foundation: MFA, conditional access policies and device compliance signals are made ready at the identity provider. In a Microsoft environment, Conditional Access planning forms the basis of this step.
- Pilot: Start with a few web-based applications and a limited user group; latency, user experience and support tickets are measured.
- Application-by-application migration: Applications are moved to ZTNA in groups; with each move, the related VPN rule is narrowed. The migration doesn’t count as “complete” until access through the VPN is removed. Otherwise you’ve only added a new door.
- Vendor access: Third-party accounts are moved to ZTNA, and their VPN accounts are closed.
- Narrowing the VPN: Remaining VPN use is documented as an exception, limited with MFA and posture checks, and reviewed periodically.
To assess the maturity of the migration, the CISA Zero Trust Maturity Model offers a useful framework with its identity, device, network, application and data pillars. The remote access side of the architecture can be handled under VPN and secure remote access, and the overall model under Zero Trust security architecture. For the same identity and device approach on the internal wired network, see the 802.1X and NAC migration plan article.
Frequently asked questions
Can ZTNA completely replace VPN?
For user-to-application access, it can in most scenarios. For site-to-site connections, server-to-client initiated connections and some legacy applications, VPN or other solutions remain necessary for a while; that’s why most organizations use a hybrid model during the transition.
Does ZTNA necessarily require a cloud service?
No. Some products allow the policy enforcement point to run in the organization’s own data center. The cloud model makes operation easier; the self-hosted model keeps traffic from passing through the vendor’s infrastructure. Data residency and operational capacity determine the choice.
What does a device posture check check?
It varies by product; typically it evaluates whether the device is managed by the organization, disk encryption, whether the EDR agent is running, the operating system version and the firewall state. These signals go directly into the access decision.
Which user group makes sense to start the migration with?
Usually a pilot group using web-based applications and vendor access are the first candidates. Moving vendor access from VPN to application-based, time-limited ZTNA access reduces risk quickly and measurably.