Most guidance on operational technology begins by telling you to keep OT off the internet, which is sound advice and no help at all if some of it is already there. The harder question is the practical one: how do you find your internet-facing OT devices when the people who created the exposure were not the people who maintain the asset register? This is a discovery problem before it is a hardening problem, and on industrial estates it has to be solved carefully, because the usual habit of scanning first and asking later can interrupt a live process.
What follows is a method you can run from a single domain, using observation rather than probing, and a set of five steps that end with a confirmed owner rather than a spreadsheet of guesses.
What Counts as an Internet-Facing OT Device
A device is internet-facing if it answers from a public IP address. Intent does not enter into it. A programmable logic controller behind a port forward on a contractor’s router is internet-facing in exactly the same sense as a deliberately published web portal, and an attacker scanning the IPv4 space does not care which one you meant to expose.
In practice the range is wider than people expect: HMIs and SCADA web portals, protocol gateways and serial-to-IP converters, remote-access appliances and engineering jump boxes, building management and access-control controllers, and telemetry units sitting on a cellular APN with a routable address. None of these is exotic. They are the ordinary sediment of an estate that has been commissioned, extended and handed between contractors over a decade.
Search results conflate OT with IoT constantly, and the distinction matters for how you respond. A compromised camera is a nuisance and a pivot point. A compromised controller can change a setpoint, trip a process or take a site offline, and the available responses are constrained by safety cases, warranty terms and the simple fact that you cannot reboot a pump station to test a theory.
The uncomfortable truth is that the asset register was written by people who did not install the exposure. The register records what was procured. It does not record what a commissioning engineer configured on a Saturday to get a plant running.
How OT Ends Up Exposed Without Anyone Deciding
Nobody signs off an exposed HMI. It accumulates, by a small number of very predictable routes.
Vendor remote support is the most common. A maintenance contract requires an inbound path so the integrator can diagnose faults without a four-hour drive, the path is created during the project, and it is still live three contracts later. Then there are convenience port forwards: a commissioning weekend, a deadline, a router owned by the facilities contractor rather than by corporate IT, and a rule that was always going to be reversed on Monday. Cloud SCADA and monitoring portals arrive on subdomains nobody in IT provisioned, often registered by the integrator on your domain at your request, long since forgotten. Flat networks contribute their share, where a segment described as segregated turns out to share an edge device with corporate traffic. And mergers, new sites and third-party operators bring their own public IP space into your name, usually with no handover of what is in it.
Consider a water or facilities site where the building management system was commissioned by a contractor who added a port forward on the site router for out-of-hours support. The corporate asset register will never show it, because the router is not corporate kit and the BMS was bought as part of a mechanical and electrical package. Five years later, that path is still open and the supplier who created it has been replaced twice.
That is the pattern worth internalising: the most common OT exposure is not an unpatched CVE. It is an inbound path someone created on purpose for maintenance and never removed. Patching advice misses it entirely, because there is nothing to patch.
Five Steps to Find Your Internet-Facing OT Devices
Each step builds on the one before it. The order matters, and the last one is the one people skip.
Step 1: Set scope and agree the rules before you look
Write down your seed data first: the domains you know you own, the registered netblocks, the site list, and the names of integrators and operators who run anything on your behalf. Then agree the rules of engagement with the engineering team in writing. Passive observation by default. No active probing of anything that might be a controller without a named owner, a change record and an agreed window. Put that in the document now, because the pressure to just check something will arrive later, usually at 4pm on a Friday.
This is also the point to decide who you call when you find something. Discovery that produces findings with no route to an owner becomes a backlog, and a backlog becomes an audit observation.
Step 2: Expand from one domain using certificate transparency
Certificate transparency logs are unusually productive on OT estates. Every publicly trusted certificate issued since 2018 is logged, and integrators and cloud SCADA platforms name things exactly as you would hope: scada., hmi., bms., plc-, telemetry., plus site codes and plant references. Those names are published even when the service behind them sits behind authentication and appears in no internal register anywhere.
Start with one apex domain, pull every certificate issued for it and its subdomains, and read the list as a map of projects rather than a list of hosts. A name like bms-stoke.example.co.uk tells you a site, a system type and, from the issuer and validity dates, roughly when someone commissioned it. The same approach underpins most of the free techniques for surfacing forgotten internet-facing assets, and it costs nothing but time.
Step 3: Work the DNS, WHOIS and netblock trail
Resolve everything you found. Note the addresses, then look at what else lives in the same netblock and what reverse DNS says about it. Site-level ranges give themselves away: a /29 from a regional ISP with three records all referencing the same plant, or a static address on a mobile operator’s business APN serving a telemetry outstation.
WHOIS and registration data then tell you which ranges and domains genuinely sit under your organisation and which belong to a supplier. Keep both. Records that resolve to nothing deserve a separate column, because a CNAME pointing at a deleted cloud resource is a finding in its own right rather than a dead end.
Step 4: Follow the supplier and operator names
Plenty of your exposure is not registered to you at all. Run the same expansion against your integrators, your facilities management provider and any operator running a site on your behalf, looking for names that carry your site codes, your brand or your plant references on their infrastructure. A dashboard called client-portal-yourname.integrator.co.uk is your risk whatever the WHOIS says, because it is your process data and, when it fails, your regulator.
Mapping who actually owns each third-party subdomain is tedious and worth doing properly, because ownership determines who can fix the thing and how long it will take.
Step 5: Confirm attribution before anyone changes anything
Discovery produces candidates, not an inventory. On OT that distinction is a safety control. Blocking traffic or pulling a route on the strength of an unconfirmed guess can interrupt a live process, and the recovery is never as quick as the decision was.
Build an evidence stack for each candidate: registration data, certificate subject and issuer, hosting netblock, shared infrastructure fingerprints, and finally a human being who recognises it. A typical confirmation chain runs like this. A certificate transparency entry shows hmi-north.example.co.uk. WHOIS matches your registrant details. The address traces to a netblock serving a regional site. The site engineer picks up the phone and identifies it as a vendor HMI installed four years ago for a line that is still running.
Banners and fingerprints help you infer vendor and protocol without touching the device, but hold that lightly. A response header suggesting a particular gateway product is an inference, not a confirmed fact. Legacy controllers can behave badly under malformed or simply unexpected traffic, and some will fault on a port scan that an IT server would not notice. That is why the general EASM advice to follow up passive discovery with active verification needs an explicit, agreed test window on OT rather than being left on by default. The difference between likely and definitely is the whole reason this step exists.
The External Issues That Matter Beyond Missing Patches
Once the list is confirmed, resist the urge to sort it by CVE count. On OT the configuration problem is usually the faster route in, and patching is frequently blocked by warranty terms, vendor validation or the absence of any maintenance window before the autumn shutdown.
Look first for administrative and engineering interfaces reachable without a VPN, including web consoles parked on non-standard ports in the belief that this constitutes concealment. Then default or shared credentials on remote-access appliances, and legacy management protocols still enabled because disabling them was never tested. Certificate hygiene matters twice over: an expired or self-signed certificate on a plant portal is a weakness, and a wildcard or a crowded subject alternative name field quietly publishes the structure of your estate to anyone reading the logs.
Dangling DNS belongs on this list too, and it is a live OT problem rather than a web one. A retired cloud SCADA pilot leaves telemetry.example.co.uk pointing at a deleted cloud resource; whoever registers that resource name next controls a hostname that carries your branding and your operators’ trust. The mechanics are the same as any other subdomain takeover through a dangling record, but the consequence is different when the name was printed in a plant handbook.
Keeping Watch Without Drowning the Team
New exposure does not appear on a schedule that suits quarterly reviews. It appears on a commissioning day, during a contractor visit, or the week a new site joins the group. Change detection beats a point-in-time look for exactly that reason, and the useful alert is a narrow one: a new name, a newly reachable service, a certificate issued on a subdomain nobody requested. Restating the known estate every week trains people to ignore the report.
Tier the estate honestly. A safety-related or production-critical controller is not the same finding as a car park barrier, and treating them identically guarantees the important one waits behind the trivial one. Our guide to scan frequency by asset tier sets out that logic in more detail.
Be clear about what external discovery does not do. It sees your estate the way an attacker does: from the outside. It does not see inside the OT network, it does not replace segmentation, protocol-aware internal monitoring or authenticated vulnerability management, and it will not tell you whether a controller is patched. It tells you what is reachable and what has changed. That is the whole discipline in one sentence, and it sits alongside the rest rather than above it.
A Short Remediation Order of Play
- Immediately: remove unauthorised inbound paths, close stale port forwards, and retire orphaned remote-access appliances whose contract ended years ago. Confirm the owner first, every time.
- Next: move vendor support behind brokered access with per-session approval and logging, rather than a permanent open port maintained on trust.
- Then: clean the dangling DNS records, replace expired and self-signed certificates on plant portals, and record each confirmed asset against consistent fields: site, system type, owner, supplier, inbound paths and criticality tier.
- Governance: write exposure terms into supplier contracts, require notification of any inbound path, and recheck externally after every commissioning or site change. A clause nobody verifies is decoration.
Regulatory pressure is moving the same way. Operators of essential services already sit under the NIS Regulations 2018, the Cyber Security and Resilience Bill was introduced to Parliament in 2025 and is still progressing at the time of writing, and the IEC 62443 series remains the reference engineers recognise for industrial automation security. The NCSC’s guidance on securing remote access to operational technology makes the same point this article does, in fewer words: know every path in, and justify each one.
The Short Version
You cannot secure a controller you do not know is reachable. Start from one domain, expand through certificate transparency, DNS, WHOIS and netblocks, extend the search to your integrators, and treat everything you find as a candidate until an owner confirms it. Keep discovery passive on OT by default and agree any active testing with engineering in advance. Then fix the inbound paths before you argue about patches, and watch for new names rather than rereading the old ones. If the term itself is new to you, start with what external attack surface management actually is and work forward from there.
Frequently Asked Questions
What counts as an internet-facing OT device?
Any operational technology that answers from a public IP address, whether or not anyone intended it to. That includes HMIs and SCADA portals, protocol gateways, remote-access appliances, BMS and access-control controllers, and cellular telemetry units with routable addresses. Exposure created by a port forward on a supplier-managed router counts exactly as much as a deliberately published service.
How do OT devices end up on the public internet in the first place?
Usually through maintenance, not negligence. A vendor needs remote support access, a commissioning engineer adds a temporary port forward, a cloud monitoring portal is registered on a subdomain IT never provisioned, or an acquisition brings its own public IP space with no handover. Each decision made sense at the time and none of them were reversed.
Can I scan internet-facing OT devices safely, or will active testing break them?
Treat active probing of OT as a change, not a check. Some legacy controllers fault on malformed or unexpected traffic, and a scan that an IT server shrugs off can disrupt an industrial device. Stay passive by default, use certificates, DNS and banners for inference, and agree an explicit test window with engineering before anything is actively touched.
Who is responsible when a supplier owns the exposed device?
Contractually the supplier may own the kit and the connection, but the operational and regulatory consequence lands on you. Record third-party-owned assets on your own risk register, name an internal owner for each one, and put notification and removal obligations for inbound paths into the contract. Ownership determines who performs the fix; it does not transfer the risk.
How often should we recheck our external OT exposure?
Continuously for change, rather than periodically for completeness. New internet-facing OT devices appear on commissioning days and contractor visits, so the useful signal is a new name or a newly reachable service, reviewed as it appears. Pair that with a deeper confirmation pass after any site change, acquisition or supplier handover.
