From implicit trust to verifiable control: Zero Trust through SSE in practice

What it delivers, how to build it in two streams, and why the real gains only begin once the VPN is gone.

Many organisations still secure their environment like a castle with a moat. Anyone who has made it inside is trusted. That is implicit trust, and it worked for as long as everything and everyone sat behind the same wall.

That situation no longer exists. Applications run in the cloud and as SaaS, employees work at home, at the office and on the road, and suppliers have access to parts of the environment. Meanwhile, attackers increasingly walk in with valid credentials, harvested through phishing or infostealers. A single compromised account is then enough to move through the network unseen.

Zero Trust in practice

Zero Trust turns that logic around. Access is no longer granted on the basis of where someone happens to be, but on explicit verification: who is the user, how healthy is the device and does the context fit the request? That check is made on every access request, not once at login.

In practice this takes the form of SSE, Security Service Edge. It brings ZTNA, SWG, CASB and DLP together in one layer that applies the same identity and the same policies everywhere. In this article we set out what that approach delivers, how to build it step by step and which pitfalls we come across most often in the field.

What it delivers
  • Less impact when an incident occurs. With least privilege access, a compromised account stays confined to a small part of the environment. Lateral movement becomes structurally harder.
  • Visibility and control. You can see who uses which application, which SaaS and AI tools are in circulation and which data leaves the organisation.
  • The same security everywhere. The same policies apply at the office, at home and on the road, without traffic first having to pass through a central data centre.
  • A better user experience. No VPN client that has to be started up, or that drops the connection halfway through the day. Access works straight away, based on who you are.
  • Compliance you can demonstrate. NIS2 names access control and multi-factor authentication explicitly among its measures, and since 15 August 2026 the directive has been Dutch law in the form of the Cyberbeveiligingswet. With verifiable control over access and data, you can show that those measures actually work.
Zero Trust does not come in a box

The most stubborn misconception is that Zero Trust is a product. It is a strategy, and you give it shape with several building blocks. SSE is the layer where access decisions are enforced, but those decisions are only as good as the signals that feed into them. Four components are indispensable.

Identity provider (IdP). Identity is the foundation under every access decision. The IdP takes care of authentication, MFA and group membership. An identity governance solution keeps roles and entitlements up to date as people join, move and leave. Without clean identity data you cannot build a least privilege policy that works.

EDR. The state of an endpoint is a strong signal. EDR shows whether a device is healthy or under attack. A device with an active threat immediately gets less access, or none at all.

MDM. MDM tells you whether a device is managed and meets the standard: encryption, patch level and configuration. That is how you tell a managed company laptop from an unknown personal device.

SIEM. Logs from SSE, identity and endpoints come together in the SIEM. There they are correlated, unusual behaviour stands out and the security team can act quickly.

Together they form one chain. The IdP establishes who you are, EDR and MDM establish how trustworthy your device is, SSE enforces the access and the SIEM watches over the whole. If one link is missing, Zero Trust remains half a solution.

The approach: two streams running side by side

Zero Trust is not something you deliver in one big project. We work with two streams that run in parallel and share the same identity and policies.

  1. ZTNA governs access to your private applications, whether they sit in your own data centre or in the cloud.
  2. SWG, CASB, DLP and AI control govern everything employees do on the internet, in SaaS and with AI.

Because both streams start at the same time, all traffic falls under the same identity-driven control from the very beginning. You do not have to wait until the VPN has been retired to get a grip on SaaS and AI.

Stream 1: ZTNA for private applications

Phase 1: retiring the VPN

The logical first step is to replace the VPN with ZTNA. Users no longer get access to a network but to applications, on the basis of identity and device posture. That pays off quickly: a better user experience, a smaller attack surface at the edge of the network and a first form of explicit verification.

What works in this phase:

  • Get phishing-resistant MFA and clean identity data in place first, with accurate groups and a watertight process for joiners, movers and leavers. Every later decision depends on it.

  • Start with a pilot group, for example IT and a number of key users.

  • Begin with the applications that are used most. A few dozen applications usually account for the bulk of the traffic. From there, work your way through the crown jewels, from the most critical applications to the less critical ones.

  • Let policies run in monitor mode before you start enforcing them.

Phase 2: least privilege with private app segments

This is where the real value lies, and it is also where things most often go wrong in practice. Phase 1 replaces the front door. Phase 2 decides what someone can reach behind it. Organisations that stop after phase 1 often end up with a VPN under a fresh coat of paint: everyone can still get to everything, only the route is different.

Least privilege at application level is the opposite of a flat network. Each role sees only the applications, ports and protocols it needs for the job. If an account is compromised, the blast radius stays small and an attacker cannot look around for the next step. That makes phase 2 at least as important as phase 1.

This is how you go about it:

  1. Use the logs from phase 1 to establish who actually uses which application.

  2. Map the dependencies, including legacy traffic such as server-initiated connections and older protocols.

  3. Create an app segment per application with specific ports and protocols, without broad ranges.

  4. Link segments to roles and groups from the IdP, not to individual users.

  5. Appoint an owner for each segment who can justify who needs access.

  6. Tighten step by step, from loose to strict, on the basis of data.

Stream 2: SWG, CASB and DLP

This stream does not wait for the VPN to disappear. You start it in parallel with ZTNA, so that internet traffic and SaaS use fall under the same control from the outset.

Step 1: SWG. Employees’ internet traffic runs through a Secure Web Gateway. You get a view of what is happening and you block known risks, such as malicious categories, command-and-control destinations and risky downloads.

Step 2: CASB. CASB gives you full insight into your corporate SaaS: who uses which application, what data it holds and how that data is shared. You protect sanctioned applications with targeted policies and you restrict the use of unsanctioned SaaS in a controlled way. Shadow IT becomes visible and manageable.

Step 3: DLP, starting small. DLP is where organisations stumble soonest. Choose a limited number of data types that really matter, such as personal data or financial data, and let the policies do nothing but monitor at first. Tune the rules on the basis of the alerts. Then move step by step to warning, and only at the very end to blocking.

What works in this stream:

  • Start with visibility. Many organisations are taken aback by the number of SaaS applications that turn out to be in use.
  • Decide together with the business which SaaS is sanctioned, and offer a good alternative before you block anything.
  • Block first what is beyond dispute, such as known malware and phishing destinations.
AI: control without a ban

Every organisation is wrestling with AI at the moment. Banning or blocking it feels like the safe choice, but it rarely works. Employees turn to personal devices, personal accounts and browser extensions, and at that point you have lost sight of what happens. Besides, AI delivers real value and has become a fixed part of the working day. Shadow AI is the new shadow IT, with one important difference: your data ends up directly in a model.

The Zero Trust answer is therefore verifiable control, built up in three layers.

  1. Visibility and policy. Map which AI tools are being used and decide which ones are sanctioned, with business accounts and clear agreements about data. Offer employees a good alternative and restrict the rest in a controlled way.
  2. AI security and guardrails. Real-time protection while people chat and enter data. Sensitive information is filtered out before it reaches the model. Attempts at prompt injection and jailbreaking are blocked and unwanted output is stopped, based on recognised standards such as the OWASP Top 10 for LLM Applications.
  3. Control over AI agents. Agents increasingly carry out actions on their own, for example through the Model Context Protocol (MCP). You inspect and regulate their behaviour, so that they do not take harmful or unauthorised actions. Treat an agent as a non-human identity: least privilege, explicitly permitted actions and full logging.
Eight pitfalls we encounter in practice
  1. Treating Zero Trust as a product purchase. Technology is a means. Without policy, identity hygiene and integration with IdP, EDR, MDM and SIEM, it becomes an expensive replacement for what you already had.
  2. A lift-and-shift of the VPN. All access is carried over as one large segment in which everything is allowed. It feels like progress, but the risk has not become any smaller.
  3. Declaring victory after phase 1. The VPN is gone, the project is closed and phase 2 drops off the roadmap. This is the biggest pitfall of all.
  4. No insight into the applications. If you do not know which applications, ports and dependencies there are, you cannot build good segments. Legacy traffic then causes surprises after go-live.
  5. Policies without an owner. Without operational ownership, segments slowly grow back into broad access and temporary exceptions become permanent.
  6. Blocking without an alternative. Shut off unsanctioned SaaS or AI without a good replacement and you send employees to workarounds, often on personal devices.
  7. DLP too big and too fast. Broad rules in blocking mode produce a stream of false positives and cost you buy-in. Start narrow and learn first.
  8. Agents with rights that are too broad. An agent that can reach everything is lateral movement in a new guise.

The same principle applies to every policy: visibility first, enforcement second.

When are you finished with Zero Trust?

There is no certificate that confirms it, but you can put it to the test. If you can answer yes to the questions below, you are at a verifiable level.

Stream 1: ZTNA

  • Has the VPN been retired, and does all private access run through identity-based access?
  • Have all broad any-any segments disappeared?
  • Does every app segment have an owner and a justification?
  • Can you demonstrate what a compromised account can reach, and is that limited?

Stream 2: SWG, CASB, DLP and AI control

  • Does the internet traffic of all users run through the SWG, wherever they are?
  • Do you have full insight into your corporate SaaS, and is the use of unsanctioned SaaS restricted?
  • Is DLP active on the data that really matters, with policies that have been tuned and are being maintained?
  • Do you know which AI tools are in use, and do guardrails and DLP check the prompts and uploads?
  • Have the rights and actions of AI agents been explicitly defined, bounded and logged?

Integrations and operations

  • Does identity from your IdP weigh in every access decision, with current roles and groups?
  • Does the EDR and MDM status of a device help determine what access it gets?
  • Do all logs come together in your SIEM, and is it clear who takes action when an alert comes in?
  • Are rights reviewed periodically, and do exceptions expire automatically?

It is those last questions that tell you the most. In the end it comes down to one thing: replacing assumptions with evidence.

Start small, with the end in mind

You build Zero Trust in phases and in two streams, with a visible result after every step and in step with identity, endpoints and monitoring. The difference lies in the execution: getting from strategy to a working solution, in an environment where people simply have to be able to keep doing their jobs. That works best with a partner who does not only advise but also helps to build.

Heimdallr guides organisations from implicit trust to verifiable control. We walk the route with you, from the first step to a working configuration. And we are not doing this for the first time: we have already led several organisations through a successful Zero Trust implementation.

Curious what this means for your organisation? Contact us or send an email to info@heimdallr.nl.

Heimdallr. Your digital watchguard for what matters most.