
Fundamentals · 4 min read
Vendor Remote Access in OT: Where It Should Terminate and Who Should Decide
Vendor remote sessions should land on a broker in the industrial DMZ, never on the control network. The plant decides who connects, when, and for how long.
By Cesar GiraldoOT Principal Architect & Technical Lead, Potenza Services
Vendor remote access should terminate on a broker or jump host in the industrial DMZ, and nowhere else. The vendor's engineer connects to that host, and from there, under a session the plant opened and can close, reaches the one device they came to work on. The control network itself never accepts an inbound connection from outside. That is the rule in both ISA/IEC 62443 and NIST SP 800-82, and it is the rule most plants we walk through do not meet.
The second question, who decides, has an equally short answer: the plant, through a named owner, with a written policy per vendor. Not the vendor's support tool, not a standing VPN account that was created at commissioning and never reviewed. If the plant cannot list which vendors have a path in, how each one lands and who approves a session, remote access is not governed. It is only installed.
How remote access usually arrives
Nobody designs a flat remote access path on purpose. It accumulates. The drive vendor installs a cellular modem in the cabinet during commissioning so they can tune from the office. The integrator sets up a VPN to the engineering workstation because the project was late. The kiln control upgrade comes with its own cloud agent. Each one is reasonable on its own day, each one is documented in a different place, and each one terminates on the control network because that is where the equipment is.
Years later the plant has several paths in, owned by several companies, most of them still working. The topology drawing from commissioning shows none of them. This is the single most common finding in the topology assessments we deliver, and it is usually the first item on the roadmap.
Why the control network cannot be the landing point
A session that lands on an HMI or an engineering workstation has the rights of that machine. From an engineering workstation the vendor's laptop, and anything on it, can reach every controller the workstation can program. The plant has no way to limit the session to the one drive that needs attention, no way to see what else was touched, and no way to end it other than asking.
The industrial DMZ exists to break that chain. A broker in the DMZ terminates the vendor's connection on one side and opens a separate, plant-controlled connection to the target device on the other. The session can be limited to one device and one protocol, recorded, time-boxed and ended from the plant. The vendor still gets the work done. The plant gets back the decision.
This is a specific case of the point in our earlier post on IT versus OT security: the corporate VPN and the enterprise firewall do not extend to the plant floor on their own. A vendor VPN that terminates inside the control network is the enterprise perimeter with the plant on the wrong side of it.
What a governed path looks like
The architecture is the easy part to describe. The vendor connects to a broker in the DMZ with their own named account and a second factor. The plant approves the session, either in advance for scheduled work or live for a fault call, and the approval names the device, the window and the purpose. The broker opens the path to that device only. The session is logged, ideally recorded, and closes at the end of the window whether or not the engineer remembers to disconnect. The plant can cut it at any moment.
The policy is the harder part, because it has to be written per vendor and kept current. For each vendor with a path in: which devices they may reach, who at the plant approves, what hours are allowed, how emergencies are handled, and what happens to the account when the support contract changes hands. That policy belongs to the plant, not to the vendor's support tool.
Where the cellular modems go
The honest answer is that some of them stay for a while. A modem in a cabinet feeding a vendor's cloud portal cannot be routed through an on-premises DMZ without the vendor's cooperation and sometimes a product change. The order of work is to find every one of them, which is what the physical walkdown in an assessment is for, decide which are still needed, remove the rest, and bring the remaining ones under the same policy: named sessions, plant approval, a known end date. A modem the plant knows about and has agreed to is governed. One nobody can account for is not.
Who should own this
Remote access governance is where the question of vendor independence stops being abstract. The party writing the policy has to be able to refuse a vendor's preferred tool, require a vendor to come through the broker, and switch a vendor's access off at renewal. That is difficult for a vendor to do to itself and impossible for it to do to a competitor on the same network. It is a plant role, held by a named person or by an independent service owner acting for the plant.
If you want one practical step this quarter, ask for the list. Every vendor with a remote path into any plant, how it terminates, who approved it and when it was last reviewed. If the list cannot be produced, the assessment that produces it is the place to start.
TagsRemote AccessiDMZPurdue Model
Ready to see your OT network as it actually is?
A per-plant OT Topology Assessment returns documented topology — asset inventory, network segmentation evidence, and security-level capability mapping — delivered to a named owner on your team.