A vendor service account is usually the weakest point of access to an industrial network. It is created in a service contract rather than in a security policy: it stays active for weeks, has no owner on the plant side and leads straight to engineering workstations and controllers. PAM for OT (privileged access management in an operational technology environment) turns that standing access into five decisions made for every session: who approves it, who watches it, which protocols it uses, where the connection is initiated and how long the record is kept.
The basics — VPN versus bastion, TeamViewer-type tools, MFA — are covered in our article on secure remote access to SCADA. Here we deal with the layer above: privileged access by people you do not employ.
Why is vendor access the biggest gap in OT security?
Because it is the only path to the controllers that by design crosses segmentation: the vendor has to get inside the OT zone, and usually does so on its own terms. A production line manufacturer or a control system integrator has remote diagnostics and fast failure response written into the contract. In practice this means a cellular modem in a control cabinet, a permanent account in the corporate VPN or a single password to the engineering workstation known to the vendor’s entire service team.
Every one of these channels has the same flaw: the plant does not know who exactly uses it. When a technician leaves the vendor’s company, nobody at the plant finds out. The vendor is in practice an administrator you do not employ, do not train and cannot dismiss.
CISA, together with the FBI, the EPA and the US Department of Energy, lists securing remote access to OT among the primary actions in the fact sheet “Primary Mitigations to Reduce Cyber Threats to Operational Technology” (6 May 2025) and recommends re-evaluating the trade-offs accepted for remote access. So the question is not whether the vendor should have access, but whether access is the plant’s decision or the vendor’s habit.
How does privileged access in OT differ from administrator access in IT?
It differs in the consequences of an error: in IT an administrator’s mistake usually ends in a service outage, in OT it can stop a physical process, damage equipment or create a hazard to people. Three differences follow from that.
Process continuity. An industrial network has no maintenance window on demand. A change in a controller is safe only in a specific state of the line, and it is operations, not the IT department, that knows that state. NIST SP 800-82 Rev. 3 (“Guide to Operational Technology (OT) Security”, September 2023) says it plainly: remote access must never bypass or weaken either functional safety or cybersecurity safeguards.
System age. According to the same publication, an OT system can remain in service for more than twenty years. A workstation with an unsupported operating system will not accept an agent, and a controller has no concept of a named account. Control therefore has to work on the path to the system, as a session intermediary.
A different definition of privilege. In OT, anyone who can change the state of the process is privileged: a setpoint on an HMI panel, the project in a controller, a drive’s operation from SCADA. An operator account without Windows administrator rights can be more dangerous than the administrator of a file server, so the inventory is built around what an account can do to the process, not around groups in the user directory.
Which accounts and protocols need to be brought under control in an industrial network?
All of those through which the state of the process or the configuration of the control system can be changed — regardless of whether they formally hold administrator rights. In a typical plant they fall into five groups:
| Access point | Typical protocol | What the vendor does there | What to watch for |
|---|---|---|---|
| Engineering workstation | RDP, VNC | uploads the controller project | the uploaded project file changes process logic |
| SCADA server, historian | RDP, SSH | diagnostics, updates | a shared “service” account used by many people |
| HMI panel | VNC | viewing and changing setpoints | VNC without a password or with a default one |
| OT network devices | SSH, Telnet, HTTP | configuring switches and firewalls | Telnet sends the password in plaintext |
| Controllers | industrial protocols, e.g. Modbus TCP | reading registers, writing values | no authentication in the protocol itself |
The last row is the hardest. NIST SP 800-82 Rev. 3 notes that many OT protocols offer no authentication whatsoever — whoever has network access to a controller can write to it. Establishing who sent a command is then possible only at the intermediate point that the traffic must pass through. Add the local device accounts to the list as well, often a single “admin” with the password from the manual.
Who should approve a service technician’s session and how does the four-eyes principle work in OT?
The session is approved by the process owner — the maintenance manager or the engineer on shift — and not by the IT department, because only they know whether the line can accept a change at that moment. The security team sets the rules and runs the tool. In OT, the four-eyes principle means that a second person takes part in any change capable of affecting the process: they approve the session before it starts or watch it live.
NIST SP 800-82 Rev. 3 (§ 3.3.7) expects a remote maintenance tool to support scheduling, authorising and auditing of the work, to verify the identity of maintenance staff, and to open the connection when the work begins and tear it down as soon as it ends. In practice: a request with a ticket number and scope, approval by operations, a time window after which access expires on its own, and a session record. It is worth tying the level of supervision to what the vendor can change:
| Type of work | Who approves | Supervision during the session |
|---|---|---|
| Reading diagnostics, downloading logs | engineer on shift | the session record is enough |
| Changing setpoints and parameters | maintenance manager | session record, notification to the shift |
| Changing controller logic, uploading a project | maintenance manager and a second engineer | live observation with the option to interrupt |
| Firmware update | maintenance manager, after a test on a spare unit | live observation, rollback plan ready before start |
The table is a starting point for a conversation with operations: the level of control follows from the risk to the process, not from trust in the vendor.
How do you give a vendor access to a system without opening inbound ports into the OT zone?
You reverse the direction of the connection: the gateway on the plant side establishes it as an outbound connection, and the vendor connects to an intermediate point in the demilitarised zone (DMZ) or in the cloud. Among its recommendations for temporary remote access, NIST SP 800-82 Rev. 3 (§ 6.2.10) lists exactly this — initiating the connection from the OT environment — alongside removing access when it is no longer required, monitoring remote activity and making operations personnel aware of planned work.
The technician logs in to the access point, authenticates with multiple factors and sees only the resources for which access has been approved. The access point is a privileged access gateway acting as a jump server. The OT zone accepts no connection from the internet, and an old controller never “sees” the vendor’s address. CISA adds that OT assets should be removed from the public internet altogether and that least privilege should apply to the specific asset and scope of work.
Two things are easy to overlook. An outbound tunnel is still a way in, so it should be active only during the approved window. And reversing the direction does not remove the need for segmentation: the gateway sits in the intermediate zone between the office and industrial networks, in line with the Purdue model, not in the middle of the control network.
What should you record during a maintenance session and how long should you keep the record?
You need to record everything that makes it possible to reconstruct a change in the state of the process: the image of a graphical session, commands typed in a console, transferred files (controller projects, firmware images) and metadata — who, when, from where, under which ticket and with whose approval. Privileged session recording is as much a diagnostic tool in OT as an evidential one: when the line behaves differently three days after a service visit, the recording shows which parameter was changed and at what time.
The most common mistake is an image without approval metadata. Such a record is a video, not evidence: you can see what happened, but you cannot show that it stayed within the agreed scope of work. The second trap is not recording file transfers — at an engineering workstation it is the uploaded project file, not mouse movements, that changes process logic.
The retention period is the organisation’s decision and has to be justified. In OT it is set by three periods, and the longest one applies:
- Time for the effect to surface — with a seasonal production cycle or rare maintenance shutdowns, the effect of a change may appear only after many months.
- The vendor’s contractual liability — as long as the warranty on the work runs, the recording is material in a possible dispute.
- Post-incident proceedings — the record has to survive until the analysis and the notifications to the authority are closed.
A recording contains the technician’s personal data, so the period must also hold up under the GDPR, and access to the archive has to be restricted just like access to the systems themselves.
How do NIS2 and the Polish Act on the National Cybersecurity System translate into control of vendor access in OT?
Directive (EU) 2022/2555 does not use the word PAM, but three points of its Article 21(2) meet precisely at the vendor service account: point (d) (supply chain security, including relationships with direct suppliers or service providers), point (i) (human resources security, access control policies and asset management) and point (j) (where appropriate, multi-factor or continuous authentication). In the Polish legal order the same areas are listed in Article 8(1)(2) of the Act on the National Cybersecurity System as amended on 23 January 2026 (Journal of Laws 2026, item 252, in force since 3 April 2026) — among others point (e) (security and continuity of the supply chain) and point (n) (access control policies).
The rules do not say how to do it, only what you must be able to demonstrate:
| Requirement | What it means for the vendor account | How you demonstrate it |
|---|---|---|
| NIS2 Art. 21(2)(d); Polish Act Art. 8(1)(2)(e) | vendor access is part of assessing the relationship with that vendor, not only a technical matter | a contract with access requirements, a list of vendors with access and their accounts |
| NIS2 Art. 21(2)(i); Polish Act Art. 8(1)(2)(n) | every account has an owner, a scope and an expiry date | a register of requests and approvals, dated access reviews |
| NIS2 Art. 21(2)(j) | the technician authenticates with multiple factors before accessing the OT zone | gateway configuration, a list of exceptions with justification |
The last column is what the auditor will ask about. How to assess the OT vendors themselves against point (d) is covered in our article on OT supply chain security.
What will PAM for OT not solve and where are other safeguards needed?
PAM controls only the sessions that pass through it — it does not see access that bypasses it. Three such paths turn up in plants regularly.
The first is physical access: a technician plugs their own laptop into the control network with a cable or inserts a USB stick into an operator workstation. This calls for entry procedures, media scanning stations and port control — we write about it in the article on service technicians’ laptops and USB drives in OT.
The second is hidden channels: a modem installed when the machine was commissioned or a remote desktop tool installed “temporarily” on an operator workstation. A PAM deployment that does not begin by finding them builds a door with a lock next to an open gate.
The third is traffic inside the zone. PAM holds the person connecting from outside accountable, but it will not protect a protocol without authentication from an infected workstation that is already in the control network. That requires segmentation, zones and conduits, and industrial traffic monitoring — in other words OT security architecture design, in which PAM is one of the components.
What does PAM for OT look like with Fudo Enterprise as an example?
Fudo Enterprise is a PAM platform that acts as an intermediary (proxy) between users and servers, without installing agents — the user connects with a native client or through a browser. We treat it as an example implementation of the pattern from the previous sections; the facts below come from the vendor’s public documentation.
Protocols. The list of supported protocols includes, among others, RDP, VNC, SSH, Telnet, Telnet 5250, HTTP and Modbus, i.e. the access points from the table above. Modbus, however, works only in proxy mode and does not support joining an ongoing session. So you cannot watch a write to a controller over Modbus live — such a session is secured by approval before it starts and a short time window.
Approval and supervision. With approval required, the user has to request access before connecting, and designated people have a set time to decide; the vendor links this feature to the four-eyes principle. For a safe (a group of resources), live supervision with a set number of supervisors can be required, and an administrator can join a session, pause it or terminate it.
Recording. The platform records the complete session traffic with metadata (playback and full-text search), registers mouse movements, keystrokes and transferred files, and the data can be timestamped, which increases its evidential value.
Connection direction. The reverse proxy opens an outbound SSH tunnel to a server outside the plant network, e.g. in the DMZ, and user traffic returns through the tunnel to internal resources. For third parties the vendor offers the browser-based Fudo ShareAccess service, which according to the documentation requires only outbound connections from the Fudo Enterprise instance.
In our assessment this is a set well suited to manufacturing and energy: an agentless intermediary, OT protocols and a reversed connection direction address three problems that block conventional tools in OT. The platform is described in detail on the Fudo Enterprise page.
How do you write PAM requirements into a service contract with a vendor?
The contract has to turn standing access into a service activated on demand — this is easier to negotiate in a new contract than to force into an existing one. A minimum set of clauses:
- Named accounts. Every technician has their own account, shared accounts are prohibited, and the vendor reports a person’s departure within an agreed deadline.
- One channel. Remote access only through the gateway designated by the plant; modems and the vendor’s own remote desktop tools are a breach of contract.
- A request before the session. A ticket number, scope of work and a time window; emergency work has a faster approval path but is not exempt from it.
- Consent to recording and supervision. The vendor accepts session recording and its interruption by the plant.
- The technician’s workstation. An up-to-date system, protection against malware, multi-factor authentication.
- Duty to inform. The vendor promptly reports an incident on its side that may affect access to the plant.
Such clauses also protect the vendor: a recording with an approved scope of work is its best defence when something goes wrong after the visit for an entirely different reason.
Where do you start implementing PAM in an industrial plant?
With an inventory of access, not with choosing a tool: first a list of vendors, their accounts and channels, and only then the gateway design. An order that works in practice:
- An inventory of vendors and channels — who has access, to what, by which route (VPN, modem, remote tool, local account) and under which contract.
- Closing hidden channels — forgotten modems and remote desktop tools switched off or brought under control.
- An approval matrix — for each process area, an owner who approves sessions and a supervision level that depends on the type of work.
- A gateway in the intermediate zone — connections initiated from the plant side, multi-factor authentication, session recording with approval metadata.
- The first vendors — start with one or two vendors with the broadest access, then add others as contracts are renewed.
If you want to take this path with a team that knows both PAM and the realities of plant maintenance, see our PAM implementation and OT security architecture design services. More on protecting industrial networks can be found in the OT/ICS security hub.
Frequently asked questions
How does PAM for OT differ from PAM in an IT environment?
In the consequences of an error and in the limits of the systems. In IT, a wrong administrator command usually ends in a service outage; in OT it can stop a physical process or damage equipment. Industrial systems stay in service for more than twenty years (NIST SP 800-82 Rev. 3, September 2023), often without the option of installing agents, and many industrial protocols have no authentication at all. That is why PAM for OT works as a session intermediary, and access approval belongs to the process owner, not the IT department.
Who should approve a vendor's remote session to OT systems?
The person responsible for the process the vendor connects to — most often the maintenance manager or the engineer on shift. The IT or security team defines the rules and runs the tool, but it does not know whether the line can accept a change at that moment. NIST SP 800-82 Rev. 3 recommends that operations personnel be aware of every planned remote activity in the OT environment.
Do you need to open inbound firewall ports for remote OT maintenance?
You do not. The pattern recommended in NIST SP 800-82 Rev. 3 is to initiate the connection from the OT environment: a gateway in the plant establishes an outbound connection to an intermediate point in the DMZ or in the cloud, and the vendor connects to that point. The OT zone then accepts no connections from the internet, and the tunnel can be activated only for the approved maintenance window.
How long should OT maintenance session recordings be retained?
The retention period is the organisation's decision and has to be justified. In OT it is set by three periods: the time after which the effect of a change can surface in the process, the vendor's contractual liability for the work performed and the needs of post-incident proceedings. A recording contains the technician's personal data, so the justification must also hold up under the GDPR.
Does NIS2 require a PAM system?
Directive (EU) 2022/2555 does not name PAM. Article 21(2), however, requires supply chain security in relations with direct suppliers (point (d)), access control policies (point (i)) and, where appropriate, multi-factor authentication (point (j)). In Poland the counterpart is Article 8(1)(2) of the amended Act on the National Cybersecurity System (Journal of Laws 2026, item 252). A vendor service account in OT is the place where these three requirements meet at once.
