Skip to content

With All Purdue Respect: Three Breaches and a Blind Spot

· 6 min read · Updated

Lire en français
With All Purdue Respect: Three Breaches and a Blind Spot

The Purdue model has been called into question since the early 2020s for three sound reasons, because the industrial internet of things, or IIoT, cloud connectivity and remote access all bypass its hierarchy, and the profession has answered with solid controls, but one path remains in what we will call here the Purdue blind spot, because it never appeared on the diagram and nobody has corrected it.

The answers usually put forward, that is IEC 62443 zones and conduits, Zero Trust, microsegmentation and unidirectional gateways, are solid and deserve to be known, but they all share one feature that is rarely noticed, since each describes a network path the model had not anticipated and each responds to it with a network control.

This article does not claim that the Purdue model is dead, and it only tries to look at what the profession has corrected in recent years, and at what nobody has corrected, which is where Trust-Buster can improve the situation.

The three breaches, without indulgence

The IIoT sensor that talks to the cloud

The model rests on a simple assumption, which is that a level only talks to its neighbours, but a connected sensor that sends its measurements straight to an online service bypasses four levels at once, and this is not a textbook case, because a compressor delivered with its built-in telemetry board, a machine whose manufacturer provides predictive maintenance from its own servers or a vibration sensor installed by a contractor with its own cellular link all arrive on site bringing their own way out, and they do not ask the hierarchy for permission since they ignore it.

Remote access, now the norm

Remote working, remote maintenance and the growing scarcity of skills have made remote access an operational necessity rather than a convenience, because a German manufacturer no longer travels for an adjustment, yet the original model had no place for it, so it was grafted on, often in a hurry and sometimes with a general-purpose remote control tool installed on a supervision workstation.

The consequences are not a matter of conjecture, since on 22 January 2026 CERT-FR published a recommendation stating that ANSSI had observed several compromises of French entities in the renewable energy generation and water management sectors, and hacktivist groups achieved them by remotely accessing management interfaces of industrial installations exposed on the internet, the most successful action having shut down a wind farm for a few hours with a financial loss for the operator. The agency specifies that these activities are most often of a low technical level and that the equipment concerned was exposed without authentication or with a default password, some industrial protocols being directly reachable from the internet without encryption, so this is a path the model had not anticipated, exploited without any particular skill, on real and recent installations.

Lateral movement, once inside

The hierarchy protects the boundaries between levels, but within a zone it protects nothing, and an attacker who has gained a foothold on a level 2 workstation moves freely towards the other systems on the same level, which is called lateral movement, because the model never said what should happen there, and this is a structural flaw rather than an oversight, since Purdue describes the vertical flow of data and not the horizontal one.

The profession's answers, and they are good ones

Industrial security deserves credit for not standing idle, because four answers have established themselves within a few years and they work.

IEC 62443-3-2 zones and conduits extend the model rather than replace it, because a zone groups assets by security requirement rather than by level, so a safety instrumented system and a supervision interface can sit at the same Purdue level while belonging to different zones, since the consequences of compromising them have nothing in common, and each zone is given a target security level, while every communication path between zones becomes a conduit that must be identified, justified and secured. The contribution seems important to us, because this mechanism makes it possible to bring into the architecture the modern flows that the hierarchy could not describe, and replication to the cloud, remote access or edge messaging become conduits as soon as a connection exists, so they are controlled.

Zero Trust modernises these conduits, since rather than a single perimeter, every flow is verified at every connection, and software microsegmentation can replace hand-configured VLANs, with policies based on cryptographic device identities rather than physical ports.

The boundary between levels 3 and 4 is hardening, and this is the most widely agreed point in the field, because the industrial demilitarised zone probably remains the most consequential control in an OT architecture and it is treated seriously, with a genuine protocol break, hardware unidirectional gateways and outbound-only connectivity.

And for equipment too old to carry an agent, inline isolation offers an elegant answer, because a hardware box placed in series on the network cable in front of the PLC inspects traffic and enforces the conduit's rules without installing anything on the legacy system and without re-architecting the network.

We have nothing to object to in any of this, because it is good work on real problems.

The Purdue blind spot

Now go back over the list and look at what these elements have in common, because the sensor that talks to the cloud is a network path, remote access is a network path, lateral movement is a network path, and microsegmentation, the bastion host, that is the jump server everyone is forced through, the unidirectional gateway and inline isolation are network controls.

Every critique levelled at the Purdue model in recent years therefore describes a communication path it had not anticipated, and every answer offered is a control placed on a communication path, and this is not a reproach, because it is consistent, since the model describes data flows and it is therefore natural that it should be criticised for missing data flows and that the response should be to control data flows.

One player in this market, incidentally, states the limitation candidly, because in its analysis of the Purdue model and Zero Trust, Brixio writes that this approach modernises the conduits and the IT-OT boundary, but that it does not change the physical topology of levels 0 to 2 and that "it does not touch the PLCs", and that is precisely the Purdue blind spot, because the person who touches the PLC uses none of these paths.

The model is not outdated, it is incomplete

The usual conclusion of articles on the subject is that Purdue is obsolete, but we do not think so, because the authoritative frameworks continue to rely on it, since IEC 62443 anchors its zones and conduits in it, NIST SP 800-82 revision 3 uses it to describe its defence in depth and the reference architectures of the major vendors derive directly from it, as we recalled in the first article, and what it is really criticised for is failing to describe paths that did not exist when it was written.

But there remains one, older than all the others, which has never appeared on the diagram and which nobody talks about, because it does not go through a port and does not carry packets, and on most industrial sites it is still the one used most often.

The US doctrine of November 2025 illustrates this ambivalent position well, because it asserts that the Purdue model, IEC 62443 and the UFC 4-010-06 criteria remain authoritative frameworks for classifying OT systems, and in the same document it chooses not to use them, preferring an abstract two-layer topology judged more flexible because it does not assign components and users to rigid levels, so the vocabulary still carries authority while the grid is starting to get in the way, and the Purdue blind spot in both of them is still the person.

Next in the series: a technician who opens a cabinet and plugs a laptop into a PLC's programming port has not crossed a single firewall, yet they have entered level 1 directly, and that is the subject of the third article.

Sources: