Zero Trust
How to understand, apply and evolve Zero Trust to protect your organisation’s systems, data and users.
Pages
Page of 18
Zero trust migration: How will I know if I can remove my VPN?

Before we begin
As your organisation makes the transition to zero trust, you may begin to question whether you still need your Always On VPN (AOVPN). An AOVPN can feel like a bottleneck in a modern network architecture, so it can be tempting for organisations to see migrating to zero trust as an excuse to remove it. But we’ve noticed a worrying trend of organisations removing an AOVPN from their network before the appropriate zero trust measures are in place.
If you take away just one thing from this blog, it should be:
When migrating to a zero trust architecture, you shouldn’t remove your VPN unless you’ve achieved the same security benefits in other ways.
Important to consider
Migrating to zero trust isn’t as simple as moving your services off-premises, making use of software as a service (SaaS) and other services, and then turning off your AOVPN. In practice, you’ll need to consider whether every part of your network can implement the zero trust principles, including any legacy systems, and also think about how your client devices communicate securely with these services.
We’ll consider here some of the ways you might do this. You might also find the suggestions in Zero trust: Building a mixed estate helpful.
Advantages of an AOVPN
In a typical organisation, data-in-transit protection for remote clients is achieved using an AOVPN across an untrusted network. An AOVPN is also often used to facilitate defence-in-depth protection of resources and services behind an IP address allow-list.
Although they are the two main benefits of an AOVPN, our guidance summarises six other advantages:
- it enables legacy systems to work remotely
- it provides a second layer of defence against misconfigured, unpatched or poorly designed internal services
- it protects internal network servers from external and unauthenticated attackers
- it protects user devices from network attacks
- it forces traffic between a device and external services through internal protective-monitoring tools
- it allows for business monitoring and filtering of users’ network traffic
But it’s important to note that an AOVPN isn’t the only security control you need to adequately protect your network. Controls such as end-user device configuration and email security are still vital components of a secure architecture.
Migrating from an AOVPN applying the zero trust principles
We’ll now look at how you can replicate some of the security benefits of an AOVPN with additional technology, and which of the zero trust principles you’ll need to consider.
Note that this isn’t an exhaustive list, and none of these technologies will replace the benefits of an AOVPN in isolation, so to put in place the right combination of solutions for your network, it’s important to understand the properties each one provides.
All communication within your network should use a secure transport protocol, such as TLS, to protect data in transit. Any insecure or inadequately protected protocols should be blocked from leaving the device. Zero trust principle 1: Know your architecture, including users, devices, services and data is critical for understanding the scope of your network and making sure that no link is missed.
Unless all communication between services in your network is protected with a secure transport protocol, you should not remove your AOVPN.
Zero trust principle 5: Authenticate and authorise everywhere means that all links within your network should be mutually authenticated, and zero trust principle 2: Know your user, service and device identities is critical to allow you to do this. Common ways to achieve this might include: user credentials (for example, using a password or multi-factor authentication) or client certificates (issued by a privately hosted PKI). More information about this is available in our enterprise authentication policy guidance.
You should note that this won’t necessarily be achieved with a secure transport protocol such as TLS which, by default, only proves the identity of a service to a client, but not the other way around.
Unless every device or service in the network is able to mutually authenticate with every other device or service it communicates with, you shouldn’t remove your AOVPN.
Although zero trust principle 8: Choose services designed for zero trust is the ideal approach when choosing new services or building a network from scratch, this may not be possible for an existing network where legacy systems are in use. Examples of legacy services might include: business productivity web applications or video conferencing.
Using managed tunnels is a popular option for accessing specific legacy services outside the network. This allows access to a few well-defined services when you can’t implement zero trust principles directly. Our VPN guidance provides more detail about managed tunnels.
Unless you have made sure that connections to all legacy systems are protected, you shouldn’t remove your AOVPN.
Devices should be configured to prevent attacks from an untrusted network. Zero trust principle 7: Don’t trust any network, including your own means that by definition all network connections are untrusted. All devices should therefore be configured to block all inbound connectivity, unless explicitly defined. More advice on this can be found in our device security guidance.
Unless every device is configured to block inbound connectivity by default, you shouldn’t remove your AOVPN.
Just as for devices, zero trust principle 7 means that all servers should also assume that all network connections are untrusted. One way to do this is to install an authenticating proxy in front of all network servers. This provides a second line of defence which ensures that unauthenticated connections can’t access network servers. It can also be dynamically adjusted to require stronger authentication if compromise is suspected, or during an incident.
Unless you have a mechanism to protect internal network servers from external and unauthenticated attackers, you shouldn’t remove your AOVPN.
Traditionally, generic web traffic would be protected in transit over the AOVPN. It would transit the VPN, and then break out to the internet at a trusted point that you control, perhaps via an on-premises proxy server. The proxy server may provide several functions, including filtering traffic based on browsing policies, blocking access to known malicious content, and monitoring and auditing of browsing.
There are several endpoint technologies that can provide the same benefit of enforcing web-filtering policies and blocking access to known malicious web content outside an AOVPN.
Even if you use one of these endpoint technologies, there is a residual risk of inadvertent data exposure or tampering as traffic leaves your device, because it won’t be protected at the first hop out to the internet. (Note that with an AOVPN, this first hop would be protected in transit, before breaking out to the internet at a location you trust, such as your office or a cloud service provider.)
Zero trust principle 7: Don’t trust any network, including your own means introducing another security control to protect data in transit at that first hop. This could be via a proxy server that supports an authenticated TLS tunnel to an SaaS proxy, guaranteeing that your data in transit is protected to the same level as a VPN, before it breaks out to the internet at a point that you can control and monitor.
An SaaS-based proxy may offer additional benefits over a traditional proxy in your on-premises architecture. For example, most SaaS proxies in the market will scale to accommodate large bandwidths and remove the bottleneck you can sometimes get with an AOVPN.
Unless traffic between devices and external services is explicitly protected for this first hop to the internet, you shouldn’t remove your AOVPN.
DNS requests have often been routed over the AOVPN, making your network less susceptible to common attacks, such as DNS hijacking. We must be able to assure that we can protect these look-ups in the same way we would with an AOVPN. DNS queries are typically sent unencrypted, so it’s important to protect them using other means.
A good DNS solution should provide the following functionality:
- protecting DNS requests in transit
- filtering out known malicious domains
- filtering out inappropriate domains
- auditing and monitoring of DNS requests
- visibility of indicators of compromise (if, for example, a DNS connection is used for malicious purposes)
- high availability
Unless you have alternative mechanisms for monitoring and filtering network traffic, you shouldn’t remove your AOVPN.
Find out more
Now we’ve looked at the different factors to consider when migrating to a zero trust architecture, it’s worth reiterating our message at the start: removing an AOVPN should be done with care. You need to consider each of the security benefits an AOVPN adds to your network, and make sure you replicate each one in your zero trust architecture.
It’s also worth noting that your network may have some properties which mean an AOVPN will always be necessary. If your network is large, complex, or contains many legacy systems, you may find that it’s not possible to achieve the same level of security in a zero trust architecture with current technology. But even when this is the case, combining your existing AOVPN with additional zero trust technologies will still improve the overall security of your network.
To read more about zero trust, have a look at our previous blogs which consider whether zero trust is right for you (now archived), and where to start with migration. And don’t forget to keep the zero trust principles themselves at the forefront of every stage of your migration journey.