Monday, production meeting at a plant in the Midwest, or in Bavaria, or in northern Italy; the story is the same everywhere. The operations director wants OEE for three lines on a tablet. Maintenance wants to predict spindle failures. A vendor of industrial IoT services arrives with a ready answer: a gateway wired into the machine network, an LTE modem, data in the cloud within two weeks. Plant IT finds out about the project when the gateway is already mounted in the cabinet next to line two. It works, data flows, everyone is happy. Nobody asked whether the same LTE modem lets someone outside talk to the controller.
Six months later the machine builder asks for remote access to the PLC to diagnose a fault. Someone creates a VPN account into the office network with a route through to the line network. The account stays, because “it will be useful next time”. The shop floor now has two entrances that appear on no diagram, and the plant IT manager is accountable for the security of a network he cannot see.
This article is for CTOs, plant IT managers and the people who connect the shop floor to the rest of the business. It defends one thesis that many IIoT platform vendors will dispute: OT security does not start with a tool; it starts with the direction of connections. Connections are opened only from a lower network into a buffer zone, no system in IT or in the cloud initiates a session into the control network, and an industrial IoT design that requires one gets redesigned, even if the vendor insists its gateway is secure. Everything else, from OT network segmentation to NIS2 compliance, is easier to build on that rule than without it.
Why is “let’s connect the machines to the cloud” the wrong question?
Because it assumes the problem is moving data. Moving data is the easy part. Protocols such as OPC UA and MQTT, edge gateways and cloud platforms are mature, and collecting machine data is now a matter of weeks, not months. The hard part is that every new connection between the shop floor and the outside world changes the risk profile of the whole plant. A network that for years was isolated, or joined to the office through one well-known point, suddenly has several new entrances, each with a different owner.
The right question is: who can open a connection into the control network, and from where, and what happens if the wrong person does. The answer drives the architecture, the vendor choice, the service contract and whether the plant meets its legal obligations. If that question is first asked after the gateway is installed, the IT team is no longer designing security; it is auditing other people’s decisions.
The second difference is cultural. In an IT project the natural goal is integration: more connections, more data, less manual work. In OT every connection is a cost that has to be justified. In its fact sheet Primary Mitigations to Reduce Cyber Threats to Operational Technology, published on 6 May 2025 with the FBI, the EPA and the Department of Energy, CISA lists removing OT connections to the public internet as the first mitigation. The reasoning is blunt: OT devices lack authentication resistant to modern threats, and attackers find them by searching public IP ranges for open ports. The agency describes the attackers’ toolsets as “simple, repeatable, and scalable”.
That is why the thesis here is not caution for its own sake. Industrial IoT does not cancel the rules of segmentation; it makes them stricter, because it multiplies the places where they can be broken. A project that starts with a connection map costs a few extra days at the beginning. A project that starts with a gateway in the cabinet costs the same days later, under pressure and with far fewer options.
How does OT security differ from IT security, and why can’t you copy the office playbook?
NIST SP 800-82 Rev. 3, published in September 2023, defines operational technology as programmable systems or devices that interact with the physical environment or manage devices that do. That is broader than classic industrial control systems: it includes building automation, transportation systems, physical access control and environmental monitoring. The common thread is that failure does not end in a data breach. It ends in a stopped line, a damaged machine or a risk to people.
That reverses the order of priorities. In the office, confidentiality dominates: customer data, financials, intellectual property. On the shop floor, availability and process safety dominate. A controller that stops responding for a few seconds because a vulnerability scanner sent it an unusual packet can halt production. Practices that are routine in IT, such as aggressive network scanning, automatic overnight patching or an EDR agent on every machine, can do more damage in OT than an attacker.
The third difference is lifespan. An office laptop is replaced after a few years. A PLC, an HMI panel or an engineering workstation beside a machine often runs for fifteen years or more, and the workstation’s operating system is the one the machine builder certified. An update may require re-commissioning and the vendor’s approval. Every plant therefore has devices that cannot be patched in any reasonable time and must be protected in other ways: segmentation, traffic control and restricted access.
The fourth difference is protocols. Many industrial protocols were designed when control networks were physically isolated and have no authentication or encryption. Whoever can send a packet to a controller can often change its state. Newer standards such as OPC UA include security mechanisms, but in a typical plant they run alongside devices that are twenty years old. In OT, “who can connect” therefore matters more than in IT, where strong authentication on every service acts as a second line of defence.
The takeaway for a CTO: you cannot move the office security policy onto the shop floor by changing the nouns. OT needs its own model, in which the main control is network architecture rather than an agent on the device. The company’s general security and compliance rules still apply, but on the shop floor it is a starting point, not the answer.
Is the Purdue model obsolete in the age of industrial IoT?
The Purdue model, derived from the Purdue Enterprise Reference Architecture developed at Purdue University, divides a plant into levels. Level 0 is the physical process: sensors and actuators. Level 1 is basic control, the PLCs. Level 2 is supervisory control: SCADA and operator panels. Level 3 is site operations: MES, historians, engineering workstations. Levels 4 and 5 are enterprise systems such as ERP and email, and the internet. Between levels 3 and 4 sits a buffer zone, often called level 3.5 or the industrial DMZ. The model is closely related to ISA-95, which we discussed in the context of the boundary between ERP and MES in manufacturing software.
The critics are right about one thing. Industrial IoT flattens the hierarchy. A vibration sensor on a spindle with its own radio does not send data to the PLC, then to SCADA, then to the historian and finally to ERP. It sends it straight to an edge gateway, which forwards it to an analytics platform. The machine builder installs a remote service module in the control cabinet that connects to its own cloud. A quality-inspection camera with an embedded vision model reports results to the quality system, bypassing MES. Anyone who tries to force these flows onto a rigid ladder of levels soon finds that the drawing no longer matches reality.
Part of the market concludes from this that Purdue is dead and zero trust replaces it: every device authenticates itself and the network stops mattering. In OT that conclusion is premature. Zero trust assumes that every participant can authenticate and can have access policies applied to it. A controller speaking an unauthenticated protocol cannot, and the plant will run it for another decade. As long as such devices are in service, network architecture remains the primary security control.
So the position of this article is: Purdue as a rigid hierarchy is obsolete; Purdue as a map of criticality and direction of flow is still essential. Flattening data flows does not mean flattening trust. An IIoT sensor can send data straight to an analytics platform, but not over a channel that lets anyone outside reach level 1. If the architecture does not guarantee that, the design is wrong, however modern it looks. Rather than arguing about levels, it helps to use a vocabulary that describes real flows better: zones and conduits.
How does IEC 62443 describe zones and conduits, and what does that mean for a project?
The ISA/IEC 62443 series is today the main reference for securing industrial automation and control systems. ISA lists its parts by role. Part 2-1 sets security programme requirements for the asset owner, which is the plant. Part 2-4 covers service providers: integrators and maintenance contractors. Part 3-2 covers security risk assessment for system design, part 3-3 system security requirements and security levels, and parts 4-1 and 4-2 the secure development lifecycle and technical requirements for components, which is what a plant can demand from device makers. Technical report 2-3 addresses patch management in control system environments.
The most practical concepts in the series are zones and conduits. CISA’s asset inventory guide, published on 13 August 2025 with partners including Germany’s BSI, the Dutch NCSC and the Canadian Centre for Cyber Security, summarises them this way: a zone is a grouping of logical or physical assets that share common security requirements based on criticality and consequence. All machines that control one production line might form one zone because they need the same protection. A conduit is the group of assets dedicated exclusively to communications through which traffic passes between zones.
What does that mean for an industrial IoT project? First, every new device must be assigned to a zone before it is connected. An edge gateway collecting line data is neither “part of the line” nor “part of IT”. It is an asset that must be placed deliberately, usually in its own zone or in the DMZ. Second, every data flow between zones must go through a named conduit, with a documented protocol, direction, purpose and owner. A flow that is not on the conduit list is, by definition, not allowed.
Third, zones differ in the level of protection they need. Part 3-3 defines security levels that describe how determined an adversary the system must withstand. Not every zone needs the highest level. A zone containing safety instrumented systems, which stop a machine when something goes wrong, needs more than a zone with information screens on the shop floor. The distinction lets you spend the budget where consequences are greatest rather than spreading it thinly across the whole site.
Fourth, zones and conduits give a shared vocabulary to three parties who usually talk past each other: the plant, the integrator and the machine builder. When the integrator’s contract says “the new line is a separate zone, connected to the DMZ by one conduit that permits only OPC UA traffic from the line to the historian replica”, there is nothing to interpret. When it says “the integrator will ensure network security”, each party reads something different.
What must sit in the industrial DMZ, and what must never be there?
The industrial DMZ is where traffic from IT ends and where traffic towards IT begins. CISA’s May 2025 fact sheet states that segmenting critical systems and introducing a demilitarised zone for passing control data to enterprise logistics reduces the potential impact of cyber threats and the risk of disruption to essential OT operations. The core rule of a DMZ is simple: no connection passes straight through it. Traffic from the office network terminates on a server in the DMZ, traffic from the shop floor terminates on another server in the DMZ, and systems on either side never talk to each other directly.
What usually sits in a well-designed DMZ? A historian replica: the plant writes process data to the level-3 historian, which replicates it to a copy in the DMZ used by reports, analytics and ERP. A message broker to which shop-floor systems publish and IT systems subscribe. An update and antivirus signature server for OT, which pulls verified packages from IT and serves them to stations on the floor. A jump host for remote access, covered below. And increasingly the cloud gateway, the only place from which shop-floor data leaves for the IIoT platform.
What must never be there? Enterprise domain accounts that grant access to systems on the plant side. If an attacker takes over an account in IT and the same account works on an engineering workstation, segmentation is meaningless. The plant needs separate identity management for OT, or at least a separate domain with no trust relationship to the office. Nor should there be services that open connections down into the control network on their own, or “temporary” firewall rules that let office traffic straight through to controllers. Every such rule needs an expiry date and an owner.
It is worth being honest about what a DMZ does not solve. It does not protect against a threat already on the plant side, such as an infected USB stick plugged into an engineering workstation. It does not protect against an integrator who configures a new line with a direct internet connection. It does not replace the ability to run manually when things go wrong. It is necessary, not sufficient. But without it every other control is easier to bypass.
Where risk is especially high, plants consider unidirectional gateways, often called data diodes, which physically prevent traffic towards the shop floor. They cost more and are less flexible than a firewall, because they rule out any command from outside, but for zones with safety functions they can be justified. The data diode decision is a good maturity test: a plant that can list every flow from the shop floor to IT also knows whether any of them needs a return path.
How do you connect an IIoT gateway and the cloud without breaking the direction rule?
Industrial IoT services, from machine data collection to predictive maintenance, need data from levels 1 and 2. In most cases they do not need the ability to send commands to those levels. That asymmetry is the foundation of a secure architecture. Data flows upwards: a PLC or an OPC UA server at level 2 passes it to the gateway, the gateway publishes to a broker in the DMZ, and a service in the DMZ sends it to the cloud over an outbound connection. No inbound connection from the cloud is needed for data to reach the platform.
The typical mistake is a gateway placed directly in the line network with its own cellular connection to the cloud. For the vendor it is convenient: no conversation with plant IT, no firewall rules to request, deployment in two weeks. For security it is an extra entrance into the control network that no plant control can see. Even if the gateway only connects outbound, compromising it gives an attacker a foothold inside a zone with nothing separating it from the controllers. Such a deployment should be rejected or redesigned; that is the concrete consequence of this article’s thesis.
What about functions that genuinely need traffic downwards? Examples include changing process setpoints remotely from headquarters, updating edge device firmware, or optimising machine parameters with a model running in the cloud. Each of these is a deliberate exception, not the platform’s default behaviour. A sensible rule: a command from the cloud lands in the DMZ as a proposal, and an operator or a system on the shop floor applies it to the controller after review. Gateway updates are downloaded by a server in the DMZ and deployed by the plant in a maintenance window. Closed-loop optimisation with no human in the loop needs its own risk analysis and usually its own zone.
For plant IT this means a short list of questions for every provider of industrial IoT services before signing. Does the platform work when the gateway sits in the DMZ rather than in the line network? Are all connections initiated from the plant side? Does the platform need any inbound connections, and if so, why? Can remote control functions be switched off? Who manages gateway certificates and keys, and how are they rotated? A vendor whose answer is “that’s not necessary, our solution is secure” should drop in the evaluation.
The decisions to make before ordering the first device, from lifetime connectivity cost to gateway identity and keys, are covered in our article on IoT consulting before you build, and the basic terms are explained in the glossary entry on the Internet of Things (IoT). The point here is narrower: the platform can be as modern as you like, but the place where it meets the shop floor must be boring, predictable and drawn on the conduit diagram.
Why is vendor and integrator remote access the biggest door into the plant?
Because it is both essential and the least controlled. The machine builder needs to diagnose a PLC fault, the integrator needs to load corrected logic, the vision system supplier needs to retrain a model. Without remote access each of these means a technician on a plane and days of downtime. So in many plants, access has accumulated over years that nobody manages: permanent VPN tunnels, modems in control cabinets, accounts shared by a vendor’s whole service team, remote desktop tools installed on engineering workstations.
CISA’s May 2025 fact sheet notes that many critical infrastructure entities, and contractors working on their behalf, make “risk-based tradeoffs” when implementing remote access to OT, and that these tradeoffs deserve careful re-evaluation. The recommendations are specific: if remote access is essential, move OT assets off the public internet onto a private network connection, use a VPN with a strong password and phishing-resistant multi-factor authentication, document and configure the solution on least privilege for the specific asset and scope of work, and disable dormant accounts. The agency also warns that misconfigurations can be introduced by system integrators, managed service providers and manufacturers in a product’s default configuration.
Well-organised vendor access looks like this. The vendor has no standing connection to the plant network. When it needs access, it submits a request describing the scope of work. The plant approves a session for a fixed time window and a specific device. The vendor connects to a jump host in the DMZ, authenticates with a named account and a second factor, and from the jump host can reach only the approved device. The session is recorded, and a maintenance engineer can watch and terminate it. When it ends, access expires on its own, with nobody needing to remember to revoke it.
It sounds like bureaucracy, but in practice it is faster than today’s chaos, because approving a session takes minutes while finding out who has the password to the old modem in the line-three cabinet takes hours. More importantly, only this model tells the plant who had access, when, and to what. Without that knowledge you can neither investigate an incident nor demonstrate supply chain oversight, which NIS2 requires.
There is also a contractual dimension. Part 2-4 of IEC 62443 sets security programme requirements for service providers, exactly the integrators and maintenance contractors in question. A plant can require in the contract that the provider meets those requirements, uses named accounts, installs no remote access tools of its own, reports incidents within a defined time and removes access for staff who leave. Before demanding this from others, it is worth checking how much about your own OT estate and its suppliers is publicly visible; our article on what attackers learn about a company before the first email shows where to look. Job adverts listing controller models and integrator case studies naming the plant are a more common source of such knowledge than any leak.
Where do you start: the asset inventory or the monitoring tool?
With the inventory. The market for OT network monitoring tools is mature, and many promise automatic device discovery. That is useful, but a tool bought before the inventory answers “what is on the network”, not “what should be on it and what is critical”. Without the second answer, an alert about a new device in a line zone has no context, and an already overloaded team gets another stream of notifications it cannot assess.
The August 2025 guide from CISA and its partners, Foundations for OT Cybersecurity: Asset Inventory Guidance for Owners and Operators, sets out five steps: define scope and objectives, identify assets and collect attributes, create a taxonomy to categorise assets, manage and collect data, and implement life-cycle management. The first step includes naming who owns the inventory, assigning roles and setting the programme’s boundaries. In the second, the guide recommends combining physical inspection with a logical survey of network and documentation data. However tempting it is, this cannot be done from a desk alone.
The guide also lists fourteen high-priority attributes to collect for each asset: active and supported communication protocols, asset criticality, asset number, role or type, hostname, IP address, logging, MAC address, manufacturer, model, operating system, physical location, ports and services, and user accounts. That list is a good starting point for a plant with no register at all. You do not need everything on day one. What matters is that criticality and location are there from the start, because they decide the order of everything else.
The third step is the taxonomy. CISA recommends classifying assets by criticality or function and grouping them into zones and conduits, citing the ISA/IEC 62443 series explicitly. In the guide’s worked examples, high-criticality assets need the most stringent measures, medium-criticality assets need robust monitoring and regular updates, and low-criticality assets need basic measures but must still be in the inventory. A finished taxonomy is also a segmentation design: the zones follow from it almost automatically.
A monitoring tool makes sense as the fourth step, once the inventory exists. Passive monitoring of OT traffic then compares reality with the register, spots devices that should not be there and flows that are not on the conduit list. That is the real value of such a tool: threat detection against a known baseline, not discovering the network from scratch. A plant that buys the tool to find out what it owns usually ends up with an expensive licence and an inventory nobody maintains.
How do you patch and maintain devices you cannot reboot?
Less often than an IT auditor would like, and more carefully than a tool vendor would like. In OT a patch is a change to a production system that needs a maintenance window, testing and often the machine builder’s approval. Technical report ISA-TR62443-2-3 is devoted precisely to patch management in control system environments, and its core message is that patching in OT is a risk-assessed process, not an automatic one.
A practical model has three tracks. The first covers assets that can be updated, such as DMZ servers, level-3 servers and IIoT gateways: keep them on a regular cycle, because they are the first target for an attacker who has come through IT. The second covers assets that can be updated only in maintenance windows after vendor validation: you need a window calendar aligned with the production plan and a backlog of pending patches with a risk rating. The third covers assets that cannot be updated at all: the only answer is compensating controls, meaning tight segmentation, allow-listed traffic, blocked removable media and monitoring.
Rating vulnerabilities in OT also needs a different lens. The same critical vulnerability in a controller inside a closed zone, reachable through one conduit with one permitted protocol, means something different from the same vulnerability in a device reachable from the office network. So vulnerability management in a plant must draw on the inventory and the zone map: priority comes from the asset’s criticality and exposure, not just the vulnerability score.
Maintenance also means the ability to recover. CISA recommends practising and maintaining the ability to operate OT systems manually, because reverting to manual control quickly is vital in the immediate aftermath of an incident, and business continuity plans, fail-safe mechanisms, software backups and standby systems should be tested routinely. A well-documented market example shows why. Norsk Hydro, the Norwegian aluminium producer, was hit by ransomware on 19 March 2019. According to the company’s own account, most business areas kept production close to normal using work-intensive workarounds and manual procedures, and the company estimated the total cost at around NOK 800 million. A plant that has never rehearsed working without its systems does not have that option.
OT backups cover more than data: PLC programs, HMI configurations, engineering tool projects, workstation images. Often the only current copy of a controller program lives in the controller itself or on the integrator’s laptop. The inventory should record where the copy of each item is, when it was last verified and who can restore it. Without that, an incident response plan is a document, not a capability.
What do NIS2, the Cyber Resilience Act and the Machinery Regulation change for manufacturers?
For a large part of industry in the EU, they turn OT security from good practice into a legal obligation. Directive (EU) 2022/2555, NIS2, lists manufacturing in Annex II as an “other critical sector”, including the manufacture of medical devices, computers, electronic and optical products, electrical equipment, machinery and equipment, motor vehicles and other transport equipment. The same annex includes food production and chemicals. Medium-sized and large companies in these sectors are generally important entities. According to the European Commission, Member States had until 17 October 2024 to transpose the directive. National laws differ in detail and timing; Poland’s amended national cybersecurity act, for example, entered into force on 3 April 2026, with registration due by 3 October 2026 and obligations to be implemented by 3 April 2027. A group with plants in several countries needs a country-by-country view.
What does it mean for the shop floor? Article 21 lists ten areas that cybersecurity risk-management measures must cover for the network and information systems an entity uses to provide its services, which includes OT where it supports production. Among them: risk analysis, incident handling, business continuity with backups and crisis management, supply chain security, security in acquisition, development and maintenance including vulnerability handling, access control and asset management, and multi-factor authentication. Each has a direct counterpart in the sections above: the inventory is asset management, the jump host is access control and MFA, the integrator contract is supply chain security, the manual-operations drill is continuity.
Article 23 sets the reporting cascade: an early warning within 24 hours of becoming aware of a significant incident, an incident notification with an initial assessment within 72 hours and a final report no later than one month after the notification. For a plant this means an event on the shop floor must be recognised as a security incident, not just a breakdown, and reach the person responsible for notification within hours. Without OT monitoring and a procedure linking maintenance with security, that deadline is unachievable.
On penalties, the directive requires maximum fines of at least EUR 10 million or 2% of total worldwide annual turnover, whichever is higher, for essential entities, and at least EUR 7 million or 1.4% for important entities. Fines are still the weakest argument for a plant’s board. The stronger one is that NIS2 describes what you have to do anyway if industrial IoT is not to increase the risk of downtime.
Two product regulations sit alongside NIS2. The Cyber Resilience Act entered into force on 10 December 2024; since 11 September 2026, manufacturers of products with digital elements must report actively exploited vulnerabilities and severe incidents, and the main obligations apply from 11 December 2027. The Machinery Regulation (EU) 2023/1230 applies from 20 January 2027 and, in Annex III point 1.1.9, requires machinery safety functions to be protected against corruption, including malicious corruption, and software interventions to be recorded. That matters beyond Europe: a US machine builder or IIoT gateway vendor selling into the EU falls under both. For plants it is good news, because you will be able to demand more from suppliers, and it pays to start in current purchase orders. For US-only sites, the CISA guidance cited throughout this article is voluntary but is the most practical baseline available.
How do you write OT security requirements into a purchase of machines or IIoT services?
The cheapest option is to buy a secure device; the most expensive is to secure an insecure one for fifteen years of service. That makes the purchase specification, not the firewall, the plant CTO’s most powerful tool. On 13 January 2025 CISA and more than a dozen partners, including the European Commission’s DG CONNECT, published Secure by Demand: Priority Considerations for Operational Technology Owners and Operators when Selecting Digital Products. It lists twelve properties an asset owner should ask manufacturers about.
They are: configuration management, logging in the baseline product, open standards, ownership, protection of data, secure by default, secure communications, secure controls, strong authentication, threat modelling, vulnerability management, and upgrade and patch tooling. Two deserve special attention. Logging in the baseline product means it records configuration changes and access without an add-on security module. Upgrade and patch tooling means a documented patch process and the ability to move to a supported operating system at no extra cost.
A purchase specification for a machine or an industrial IoT service should cover at least:
- connection architecture: all connections initiated from the plant side, no inbound connections, no vendor modems or cellular cards without the plant’s consent;
- remote access: only through the plant’s jump host, named accounts, multi-factor authentication, no vendor remote desktop tools;
- documentation: a list of ports, protocols and services, default accounts and how to change them, a data flow diagram;
- life cycle: the security support period, the delivery time for critical patches, the validation procedure for workstation OS patches;
- disclosure: an obligation to inform the plant of vulnerabilities and incidents affecting the product within a set time;
- conformity: a statement of conformity with the relevant parts of IEC 62443 (4-1 and 4-2 for manufacturers, 2-4 for integrators), or a justification of deviations.
These requirements need not shut out the market. Many manufacturers already meet them, and those who do not should say so before signing, not after installation. What matters is that maintenance, IT and procurement write the specification together. Procurement alone cannot judge an answer about threat modelling, and IT alone does not know which machine functions production cannot live without. The same logic applies to software ordered for the integration layer, such as services in the DMZ; what to require from that vendor from sprint one is covered in our article on security by design and zero trust in a software project.
What does a 90-day plan look like, and how do you assess plant maturity?
The first 90 days do not need large purchases. They need a decision on who owns OT security, and a few weeks of work on the shop floor. A suggested order: in month one, an inventory of critical zones and a list of every external connection, including vendor modems and gateways. In month two, the zone and conduit map, a review of remote access accounts and removal of dormant ones. In month three, the DMZ design, the vendor access procedure and a first manual-operations drill on one line. In parallel, confirm the plant’s regulatory status in each country where it operates, because national registration deadlines will not wait for the project.
The table below helps assess where a plant stands today and names the evidence that should exist at each level. Evidence matters more than claims: “we have segmentation” without a conduit diagram is a claim.
| Area | Baseline | Target | Evidence to ask for |
|---|---|---|---|
| Asset inventory | Device list with IP addresses | Register with criticality, location, owner and CISA’s 14 attributes | Register with the date of last on-site verification |
| Segmentation | Office and production networks separated | Zones by criticality, every flow a named conduit | Zone and conduit diagram matching firewall rules |
| Industrial DMZ | Firewall between IT and OT | No pass-through traffic, historian replica and broker in DMZ, separate OT identities | Firewall rules with no direct IT-to-OT connections |
| Industrial IoT | Vendor gateways in line networks | Gateways in DMZ or own zone, outbound-only connections | Gateway list with connection direction and owner |
| Remote access | Password VPN for vendors | Jump host, requested sessions, MFA, recording, automatic expiry | Session log for the last 30 days |
| Vulnerabilities and patching | Updates “when there is a chance” | Three tracks: regular cycle, maintenance windows, compensating controls | Patch backlog with risk ratings |
| Continuity | Data backups in IT | Backups of PLC programs and configurations, rehearsed manual operation | Report from the latest drill |
| Compliance | Awareness of NIS2 | Registration where required, Article 21 measures mapped to OT assets, 24/72-hour reporting procedure | Registration record and incident reporting procedure |
| Procurement | No security requirements | Secure by Demand and IEC 62443 requirements in every specification | The latest machine purchase specification |
A plant that reaches the target level in the first three rows for its critical zones within 90 days has a foundation on which industrial IoT can grow safely. A plant that started by buying a monitoring tool usually has a handsome dashboard and still does not know which of the devices on it are critical. The gaps visible on the shop floor usually show up in IT as well, so it pays to run the review on both sides of the DMZ at the same time.
What are the key takeaways for the CTO and plant IT?
Industrial IoT gives production real value: near-real-time data on machine condition, quality and energy use. That value does not require opening the control network to the world. It requires an architecture in which data flows upwards through a buffer zone and every connection downwards is a deliberate, documented and controlled exception.
Remember six things. First, start with the direction of connections, not a tool: no system in IT or the cloud opens a session into the control network. Second, treat the Purdue model as a map of criticality, and use IEC 62443 zones and conduits as the language of design and contracts. Third, build a DMZ that nothing passes straight through, and put IIoT gateways in it. Fourth, replace standing vendor access with requested sessions through a jump host, with MFA and recording. Fifth, build the inventory before buying a monitoring tool, and rehearse manual operation. Sixth, check each site’s regulatory status under national NIS2 law, and write supplier requirements into every new specification, ahead of the CRA and Machinery Regulation dates.
IEC 62443 conformity assessment and testing on live installations are work for a specialist OT security team. But the layer between the shop floor and the business, meaning the DMZ, the message broker, data replication, integration with MES and ERP, infrastructure automation and access management, is built by IT teams, and that is where gaps most often appear. If you need to strengthen that layer, see how our DevOps engineers with DevSecOps and zero-trust networking experience work, or how we approach custom software development for integrating shop-floor data with business systems. You can also book a call and start with a review of your connection map, before the next gateway appears in a cabinet beside the line.