1. Intro2. Key Insight3. Roadmap4. Closing Thoughts
READ:
Frameworks 04-NOV-2026 · 12 min read

After Purdue: Practical Network Segmentation for Brownfield OT Environments

This article challenges the conventional wisdom on network segmentation in ICS/OT after the Purdue recommendations. It provides a practical, step-by-step guide for implementing effective network segmentation in brownfield OT environments.

brownfieldnetwork segmentationPurdue guidelinesOT securityICS security
DE.CM Continuous Monitoring
AI-Generated Locally
Article Details
CategoryFrameworks
Published04-NOV-2026
Read Time12 min read
AuthorNEXUS Engineering
Brownfield Constraints

What Makes Brownfield OT Networks Hard to Segment

In a brownfield plant, the network you have to secure is the one your passive monitoring shows you, not the one in the drawing set.

A brownfield OT environment is an operating facility whose control network grew over years or decades. Many vendors, integrators and project teams each added equipment to meet their own needs. A greenfield plant can design zones before the first controller is commissioned. A brownfield plant has to put boundaries around systems that are already running, often with incomplete drawings and little tolerance for downtime. These constraints, not a lack of frameworks, are the main reason segmentation projects stall.

The first constraint is protocols that have no security controls and, in many cases, cannot be routed. Modbus TCP (port 502) has no authentication: any host that can reach a controller can issue read and write function codes. DNP3 (port 20000) gained an optional Secure Authentication mechanism in IEEE 1815-2012, but adoption in installed field devices remains limited. Other protocols constrain where a boundary can even sit. PROFINET RT runs directly over Ethernet frames (EtherType 0x8892) rather than IP, so a Layer 3 firewall cannot sit between an IO controller and its devices. EtherNet/IP Class 1 implicit I/O traffic (UDP 2222) is often multicast and sensitive to timing, so a boundary placed in the wrong position can trip connection timeouts and fault the process. Serial protocols such as Allen-Bradley DF1 or Modbus RTU frequently reach the IP network through protocol gateways. Those gateways become hidden paths into a zone.

The second constraint is assets that cannot be hardened. Many brownfield HMIs and engineering workstations still run Windows versions that Microsoft no longer supports, because the control software is validated only on that platform. PLC firmware is rarely updated outside planned turnarounds. As a result, segmentation and conduit filtering are often the main compensating controls available, which is why NIST SP 800-82 Rev. 3 treats network architecture as a core defence rather than an optional extra.

The third constraint is hidden connectivity. Three patterns recur in field assessments: flat Layer 2 networks that span several production areas, engineering laptops or historian servers with a network card on both the business and control networks, and vendor remote-access paths (cellular modems, persistent VPNs, remote desktop tools) that bypass any DMZ. CISA and partner agencies documented attackers abusing these weaknesses in the Volt Typhoon advisory (AA24-038A, February 2024). That advisory described a state-sponsored actor that maintained long-term access in critical infrastructure IT environments and targeted pathways into OT. If an organisation draws zones from old drawings instead of observed traffic, these paths stay hidden.

The fourth constraint is the process itself. Continuous operations such as refining, chemicals, power generation and water treatment may allow intrusive network changes only during scheduled outages, which can be a year or more apart. Safety instrumented systems engineered under IEC 61511 need particular care. The 2017 TRITON/TRISIS incident, which targeted Schneider Electric Triconex safety controllers, showed why the SIS should be its own tightly controlled zone. The same incident shows that any change to SIS connectivity must go through formal management of change.

In practice, these constraints set the order of work. First, build a verified inventory and communication baseline using passive monitoring (SPAN or TAP feeds into an OT network monitoring tool), so the project does not depend on active scanning that fragile devices may not tolerate. Second, group assets by function and by the consequence of their compromise, and decide where boundaries belong. Third, pick enforcement points that respect protocol limits. Use a Layer 3 firewall where traffic is routable. Use VLAN separation or a managed switch ACL where real-time Layer 2 traffic must stay local. Use a unidirectional gateway where data only needs to flow out of the zone.

⚠️

The Purdue model describes functional layers: Level 0 is the physical process, Levels 1 and 2 are basic and supervisory control, Level 3 is site operations, and Levels 4 and 5 are the enterprise. It says nothing about how risky a given asset is or which flows are actually needed. Turning each level into a firewall boundary causes four common failures in brownfield plants. (1) Boundaries break real-time traffic. A Layer 3 firewall between Level 1 and Level 2 cannot pass non-IP PROFINET RT frames. It may also add latency or block the multicast that EtherNet/IP implicit I/O relies on, which can trip production faults. (2) Level-based rules ignore consequence. A water-chemistry PLC and a packaging-line PLC both sit at Level 1, but a compromise of each carries very different safety and environmental impact. ISA/IEC 62443-3-2 addresses this by grouping assets into zones according to risk and assigning each zone a target Security Level. Hierarchy alone does not determine the grouping. (3) Rules invite broad exceptions. When a strict level model blocks a legitimate flow, such as a historian collecting data, a vendor diagnostic session or a cloud analytics connector, teams often open any-to-any rules or add unmanaged dual-homed hosts. The result is less secure than a well-defined conduit would have been. (4) Rules create false assurance. A neat diagram with a Level 3.5 DMZ looks compliant even when cellular modems, USB-borne engineering files or remote-access tools bypass it. Use Purdue as a shared vocabulary for describing where assets sit. Build the actual segmentation with a 62443-3-2 risk assessment, a passively observed communication baseline, and conduits that each have a documented owner, permitted protocols and a review date.

Implementation Roadmap

A Practical Approach to Network Segmentation

▲

In brownfield plants, segmentation fails less often because of weak firewall rules than because of unknown dependencies. Examples include a historian that polls PLCs across a 'flat' network, an EtherNet/IP I/O connection using multicast on UDP 2222, or a PROFINET RT device that cannot be routed at all. This roadmap follows the risk-based zone-and-conduit method in IEC 62443-3-2 and the guidance in NIST SP 800-82 Rev. 3 (September 2023). It assumes that every change is made under Management of Change (MOC) and aligned with planned outages. Timelines assume a single mid-sized site. Multi-site programmes should run Phases 1 and 2 per site.

Phase 1: Discovery and Risk Assessment
Months 1–2

Build an evidence-based picture of what is actually on the network and how it communicates.

Start with passive discovery. Mirror traffic from core and distribution switches through SPAN ports or network TAPs into an OT-aware monitoring tool, either a commercial platform or open-source Zeek with ICS protocol parsers. Run active scanning only with operations approval, because some legacy PLCs and serial-to-Ethernet gateways are known to fault under aggressive scans. Reconcile the observed traffic against existing drawings, which are frequently out of date in brownfield sites. Then perform the initial (high-level) and detailed cyber risk assessments described in IEC 62443-3-2. Rank assets by consequence, such as safety, environmental release, or production loss, rather than by IP count.

Capture at least one full production cycle of traffic (including batch changeovers and shift handovers) so periodic flows are not missed. Build an asset inventory recording vendormodelfirmwareprotocol(s)and whether each device supports routing or needs Layer 2 adjacency. Produce a communication matrix of sourcedestinationprotocol/portfunctionand business owner for every observed flow. Flag high-risk pathssuch as remote-access sessionsdual-homed engineering workstationsand direct IT-to-Level 1 connections. Identify Safety Instrumented Systems (SIS) and document their current connectivity to the Basic Process Control System (BPCS).
Phase 2: Zone and Conduit Design
Months 3–4

Partition the system into zones by risk and function, and define tightly scoped conduits between them.

Group assets into zones that share security requirements, as set out in IEC 62443-3-2. Use the Purdue levels as a reference vocabulary, not a rigid template. A packaging line and its local HMI may belong in one zone even if they sit at different Purdue levels. Assign each zone a target Security Level (SL-T) and map it to the system requirements in IEC 62443-3-3. For conduits, match the control to the protocol. For Modbus TCP (port 502), ICS-aware firewalls with deep packet inspection can allow read function codes (for example 3 and 4) while restricting write codes (5, 6, 15, 16) to named engineering hosts. For OPC UA (default port 4840), DPI visibility is limited when the SignAndEncrypt security mode is used. Rely on certificate management, endpoint allow-lists, and placing OPC UA aggregation servers in an industrial DMZ. Plan for protocols that resist routed segmentation. PROFINET RT and EtherCAT run at Layer 2, and EtherNet/IP implicit messaging often relies on multicast. These usually call for cell/area zones protected at their boundary rather than split internally.

Draft a zone-and-conduit diagram and a matching rule set derived directly from the Phase 1 communication matrixusing default deny between zones. Design an industrial DMZ (Level 3.5) so that no direct session crosses from enterprise IT to control zones. Broker historian replicationpatch distributionand remote access through it. Define the remote-access architecture: a jump host in the DMZ with MFAsession recordingand time-bound vendor accounts. Keep the SIS in its own zone with the minimum necessary conduit to the BPCSconsistent with IEC 61511 independence expectations. Obtain sign-off from operationsprocess safetyand maintenance before any change request is raised.
Phase 3: Staged Implementation and Validation
Months 5–8

Deploy controls incrementally, starting in monitor mode, without interrupting production.

Introduce enforcement points in non-blocking (alert or learning) mode first. Compare logged denies against the communication matrix for two to four weeks to catch flows that discovery missed. Move zone by zone, starting with the lowest-consequence zones, and switch to blocking only during approved maintenance windows with a tested rollback plan. Test that segmentation behaves as designed. Confirm that permitted flows still work and that prohibited paths are actually blocked. Do not rely on simple latency or ping tools for this, because ICMP reachability says little about whether industrial protocol traffic is being correctly filtered.

Deploy firewalls or ICS-aware gateways at the DMZ and zone boundaries in monitor modethen tune rules from observed hits. Re-IP or re-VLAN only where devices support it. Document hard-coded IP dependencies in PLC logic and HMI projects before changing addressing. Cut over to enforcement per zone during planned outageswith operators and vendor support on standby. Validate with packet captures (Wireshark/Zeek) on both sides of each conduitfirewall rule hit countersand controlled negative tests from an approved test host. Update as-built drawingsthe asset inventoryand MOC records immediately after each cutover.
Phase 4: Monitor, Maintain, and Reassess
Month 9 onward

Treat segmentation as a living control rather than a one-off project.

Rule sets erode as vendors request exceptions and new equipment is added. Continuous network security monitoring at conduits lets defenders detect policy violations and adversary behaviour, such as the living-off-the-land techniques described in CISA advisory AA24-038A on Volt Typhoon (February 2024). Reassess zones whenever the system changes materially, as IEC 62443-3-2 expects.

Review firewall rules quarterlyremoving expired vendor exceptions and unused allow rules. Feed conduit logs and OT protocol alerts into the SOC with OT-specific playbooks. Re-run the risk assessment after major projectscloud or IIoT integrationsor new threat intelligence. Test the incident response plan with a tabletop exercise that assumes a zone boundary has been breached.
Open Questions for OT Defenders

Looking Ahead: The Future of OT Security in an Evolving Landscape

Recent incidents show why segmentation has to be engineered around how a plant actually works, not drawn from a diagram. Adversaries are increasingly targeting the industrial protocols themselves. The PIPEDREAM/INCONTROLLER toolkit, disclosed in 2022, included modules for OPC UA, Modbus, and CODESYS. FrostyGoop, reported by Dragos in July 2024, used Modbus TCP and was linked to a heating outage in Lviv, Ukraine. Meanwhile, IIoT sensors, cloud analytics, and vendor remote-support platforms keep creating paths that bypass the traditional Purdue hierarchy. NIST SP 800-82 Rev. 3 and IEC 62443 provide a durable framework. The harder problem is keeping zones and conduits accurate as brownfield sites change year after year.

01

Monitoring: If an attacker used only legitimate protocol functions inside an allowed conduit, such as a Modbus write from a trusted host, would your monitoring distinguish it from normal engineering activity?

02

Cloud and IIoT: How will you extend zone-and-conduit thinking to cloud and IIoT connections that originate outbound from Level 2 or Level 3? Who owns the risk for those paths?

03

Rule ownership: What process stops segmentation rules from decaying through accumulated vendor exceptions, and how often is it actually enforced?

04

Testing: Where encrypted industrial protocols such as OPC UA SignAndEncrypt reduce deep-packet visibility, which controls replace inspection at the boundary?

05

Scalability: Can your segmentation design absorb a new production line or a controller migration without a full redesign? Have you tested that assumption?

Segmentation is not finished when the last firewall goes in. It is only as current as your last asset inventory and your last risk assessment. Treat zones and conduits as operational controls, reviewed with the same discipline as safety systems, and they will keep pace with both the plant and the threat.
← Back to CyberCuriosity Let's Talk
Comments & Suggestions
Thoughts on this article? Corrections, questions, or additions — all welcome.
★ ★ ★ ★ ★ Optional — tap to rate
GDPR: Your data is processed solely to respond to your enquiry and is never shared with third parties. By submitting you consent to NEXUS Cybersecurity storing your details for this purpose only.
Sent privately — never published publicly