Secure Remote Access for OT Vendors and Integrators
Design vendor remote access to OT you can approve, watch and revoke: brokered sessions, DMZ jump hosts, MFA, time limits and firewall rules to enforce it.
Almost every industrial site depends on outside help. PLC integrators adjust logic, DCS vendors troubleshoot controllers, OEMs read diagnostics from packaged machinery, and support contracts often assume someone can connect from far away at short notice. That dependency is legitimate. The problem is how the access is usually built: a VPN account shared by a whole vendor team, a remote desktop tool installed on an HMI years ago, a cellular router inside a control cabinet that nobody on the security team knows about.
Remote access is also one of the best documented paths into control systems. In the 2015 attack on Ukrainian distribution companies, CISA's alert describes the attackers operating breakers remotely using legitimate credentials, through remote administration tools and ICS client software over VPN connections (see ICS attack case studies). In MITRE ATT&CK for ICS, External Remote Services (T0822) and Remote Services (T0886) are both Initial Access techniques. This guide describes an architecture that keeps vendor access possible while making it approved, observed, time limited and easy to cut.
Principles before products
Whatever tooling you choose, the design should satisfy a short list of properties. They line up with NIST SP 800-82 Rev. 3 (the "Remote Access" topic under Protect) and with IEC 62443-3-3 system requirements such as SR 1.13 (access via untrusted networks) and SR 2.6 (remote session termination).
- No direct inbound path into the control network. Remote sessions terminate in a DMZ, never on a Level 2 or Level 1 asset. If you are unsure where those levels are, read the Purdue model guide first.
- Named individual accounts. Every person has their own identity. No "vendor01" account shared across a support desk.
- Phishing resistant MFA at the point where the session enters your environment.
- Access is off by default. It is enabled for a specific window, for a specific person, toward specific assets, after a human on site approves it.
- Sessions are brokered and recorded. The vendor reaches a jump host or access broker, and only that system talks to the target.
- Least function on the target. Read only diagnostics should not come with the ability to download a new program.
- Operations can see and kill the session. The control room, not only the security team, should know who is connected and be able to terminate it.
Reference architecture
The pattern below works for a single plant and scales to a fleet. It assumes an industrial DMZ as described in OT network segmentation and the industrial DMZ.
Vendor laptop
| (internet)
v
[Remote access gateway: VPN or ZTNA, MFA, per user policy] <- enterprise edge or IDMZ
|
v
[Jump host / access broker in the IDMZ] <- session recording, file staging
| (conduit: firewall allows only defined protocols to defined targets)
v
[Engineering workstation or target asset, Level 2 / Level 1]
Key decisions in this design:
- The gateway authenticates people, the jump host controls what they do. The VPN or zero trust access gateway enforces MFA and device checks. The jump host enforces which OT targets and protocols are reachable, and records the session.
- The jump host is the only source allowed through the conduit firewall toward the OT targets. The vendor laptop never gets an IP route into the control network.
- Engineering software lives on the jump host or on a site engineering workstation, not on the vendor laptop. That way the project files and the tools that can write to controllers stay on assets you manage, patch and monitor.
- The IDMZ firewall pair (or the single firewall with separate interfaces, if that is what you have) blocks any traffic from the enterprise side directly into OT. Every session goes through the broker.
Commercial secure remote access products for OT package these functions together. They are useful, but the same principles apply whether you buy one or assemble the pattern from a VPN, a privileged access management tool and a hardened Windows jump host.
Enforcing the conduit
The conduit between the jump host and the control network is where the design becomes real. Write the firewall policy as an explicit allow list. Here is an nftables example for a Linux based conduit firewall; the same logic maps to any industrial firewall.
# Jump host 10.30.5.10 (IDMZ) -> engineering workstation 10.20.1.15 (Level 2)
# RDP only, and only when the "vendor_window" set contains the jump host.
nft add table inet conduit
nft add set inet conduit vendor_window '{ type ipv4_addr; flags timeout; }'
nft add chain inet conduit forward '{ type filter hook forward priority 0; policy drop; }'
nft add rule inet conduit forward ct state established,related accept
nft add rule inet conduit forward ip saddr @vendor_window ip daddr 10.20.1.15 tcp dport 3389 ct state new log prefix \"vendor-rdp \" accept
# Open a 4 hour window after the site supervisor approves the request:
nft add element inet conduit vendor_window '{ 10.30.5.10 timeout 4h }'
# Close it early:
nft delete element inet conduit vendor_window '{ 10.30.5.10 }'
The timeout on the set element means the window closes on its own even if nobody remembers to revoke it. Note that closing the window blocks new connections; to cut an established session, also terminate it on the jump host or flush the matching conntrack entries (conntrack -D -s 10.30.5.10).
If the vendor needs to reach a controller directly (for example to go online with a PLC from engineering software running on the jump host), add one rule per controller and per protocol rather than opening the whole subnet. The port lists in ICS protocol security help you write these rules precisely.
Hardening the jump host
A Windows jump host running RDP is the most common broker. Treat it as a high value asset:
- Dedicated and single purpose. No email, no browsing, no general file sharing.
- Separate identities. Vendor accounts live in a directory that is not your enterprise domain, or are local accounts managed by your privileged access tool. Never reuse enterprise credentials inside OT.
- Restrict redirection. Disable drive and clipboard redirection so files cannot be dragged in or out of the session outside the controlled transfer path. Through Group Policy these are "Do not allow drive redirection" and "Do not allow Clipboard redirection" under Remote Desktop Session Host, Device and Resource Redirection. The registry equivalent:
HKLM\SOFTWARE\Policies\Microsoft\Windows NT\Terminal Services
fDisableCdm (DWORD) = 1 ; no drive redirection
fDisableClip (DWORD) = 1 ; no clipboard redirection
MaxIdleTime (DWORD) = 900000 ; 15 minutes, in milliseconds
MaxDisconnectionTime (DWORD) = 60000 ; end disconnected sessions after 1 minute
- Application control. Allow only the engineering tools and utilities the vendor actually needs.
- Session recording. Record video or keystrokes of every session and keep the recordings with your incident response evidence retention.
- Controlled file transfer. Files the vendor brings in (firmware, patches, project files) land in a staging share in the IDMZ, are scanned and hashed, and are moved to OT by site staff. Hash verification ties directly into OT patch and vulnerability management.
The approval workflow
Technology does not help if access is always on. A workable process:
- The vendor opens a request naming the person, the asset, the purpose and the expected duration, ideally linked to a work order.
- The site owner of the process (operations or maintenance, not only IT) approves it and confirms the plant state allows the work.
- Security enables the account and the conduit window for that duration.
- The control room is told who is connected and to what.
- When the work ends, access closes automatically, and the vendor submits a short summary of changes made. Any PLC logic or configuration change is compared against the stored baseline.
This is also where IEC 62443-2-4, which covers security program requirements for IACS service providers, becomes useful in contracts: you can require integrators to follow your remote access procedures, use named accounts and report changes. See IEC 62443 explained for how the parts fit together.
Finding the access you did not approve
Most plants have remote access paths that were never designed. Hunting for them is part of the job:
- Cellular modems and routers in panels, packaged skids and remote sites. Physical walkdowns and a review of cellular data plans on purchase orders find more of these than network scans do.
- Remote desktop and support tools installed on HMIs and engineering workstations. Check installed software, running services and outbound connections.
- Outbound connections to remote support services visible in DNS and firewall logs. A passive monitoring sensor (see OT asset inventory) makes this easy to review.
A quick query over Zeek logs to spot RDP, VNC or SSH sessions into OT from anything other than the jump host:
zcat conn.*.log.gz | zeek-cut ts id.orig_h id.resp_h id.resp_p service \
| awk '$4==3389 || $4==5900 || $4==22' \
| grep -v '^[^ ]* 10\.30\.5\.10 '
CISA and partner agencies published a joint guide on securing remote access software because these legitimate tools are routinely abused by intrusion groups and ransomware operators. Treat any remote access tool you did not deploy as an incident until proven otherwise, and have a playbook ready for it in your OT incident response plan.
Detection and monitoring
Log and alert on:
- Authentications at the remote access gateway outside an approved window.
- Any connection through the conduit that does not originate from the jump host.
- Engineering protocol activity during a remote session that goes beyond the request: program downloads, mode changes, firmware updates. In ATT&CK for ICS terms that is Program Download (T0843), Change Operating Mode (T0858) and Activate Firmware Update Mode (T0800). The ATT&CK for ICS guide shows how to map detections to these.
- New accounts or group changes on the jump host.
Common mistakes
- Persistent site to vendor VPN tunnels. CISA's Ukraine alert explicitly recommends avoiding persistent vendor connections. A permanent tunnel turns the vendor's network into part of yours.
- MFA on the VPN only, with shared credentials behind it. MFA does not help attribution if everyone uses the same account on the jump host.
- Remote access straight to an HMI. HMIs are operator consoles. Remote control of a live HMI by someone who cannot see the plant is a safety issue as much as a security one.
- No way for operations to end a session. The people who see the process first need a simple, documented way to disconnect a remote user.
Remote access done this way costs a few minutes of process per session. In exchange, you know who touched which asset, when and why, which is exactly the question you will be asked after any unexpected change in the process. For the broader context of why availability and safety shape these choices, start with OT vs IT security, and for how NIST frames remote access within an OT program, see the NIST SP 800-82 overview.
References
- NIST SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security
- CISA: Cyber-Attack Against Ukrainian Critical Infrastructure (IR-ALERT-H-16-056-01)
- CISA: Primary Mitigations to Reduce Cyber Threats to Operational Technology (fact sheet, PDF)
- CISA and partners: Joint guide to securing remote access software
- ISA/IEC 62443 series of standards
- MITRE ATT&CK for ICS: External Remote Services (T0822)
- MITRE ATT&CK for ICS: Remote Services (T0886)
- CISA ICS Recommended Practices