Safeguard
Supply Chain Security•

How SCADA Systems Get Hacked and How to Secure Them

SCADA security explained: how PLCs, water plants and dams get hacked, from exposed HMIs and default passwords to Stuxnet, and how to secure them.

Safeguard Security Team
69 min read

SCADA systems and the industrial controllers beneath them get hacked in a handful of well-documented ways. Attackers find operator screens and controllers that are reachable from the internet and log in with default or weak passwords. They come in through remote access tools that were set up for convenience. Ransomware lands on the office network and forces the plant into a precautionary shutdown or manual operation. And in a small number of serious cases, attackers have compromised the engineering software that programs controllers and changed the logic itself. Almost every one of those paths depends on a design assumption that no longer holds: that the plant network is isolated and that anything able to talk to a controller is trusted.

That is the short answer. This guide is the long one. It explains what SCADA, DCS, PLCs, RTUs and HMIs are, how a water treatment plant or a dam is actually controlled, and how the Purdue model organises a plant network. It walks through the legacy protocols that carry commands without authentication (Modbus, DNP3 and IEC 60870-5-104), the scanning research on internet-exposed control systems, default and shared passwords, the 2021 Oldsmar water plant incident and the later doubts about its cause, ransomware that spread from IT into operations at real utilities, and the two cases that showed controller logic and safety systems can be changed: Stuxnet and Triton. It covers documented PLC vulnerabilities at major vendors, the risk inside engineering workstations and project files, the standards that apply (IEC 62443, the ISA standards and NERC CIP), and hardening checklists for small utilities, larger operators and the companies that build industrial products. It ends with the software supply chain inside every controller and what SBOMs change.

It is written from the attacker's point of view for defenders. For each attack you will find how it works and why, what the attacker needs, real cases, the impact, how to detect it and how to defend against it. You will not find exploit code, tool command lines or step-by-step instructions, and incidents are described technically, without attributing them to anyone. Where a tool is named, it is named as context for authorised testing.

A note on the law. Scanning, accessing or testing industrial control systems that you do not own, or have no written permission to test, is illegal in most countries, and on a live process it can also be dangerous. Everything in this guide is for protecting systems you operate, your organisation's sites with authorisation, or products you build. Test on lab equipment or a dedicated test bed, never on a running plant.

This post is part of our device security series. The companion guides on how Wi-Fi gets hacked, how routers get hacked and how laptops get hacked cover the networks and the engineering laptops that so many plant incidents start from. Our posts on fabless chip designers and the software supply chain and how a laptop is built explain where the chips and SDKs inside controllers come from.

Contents

  1. What are SCADA, DCS, PLCs, RTUs and HMIs?
  2. How is a water treatment plant or a dam controlled?
  3. What is the Purdue model?
  4. What is the attack surface of a SCADA system?
  5. What does an attacker need to attack a SCADA system?
  6. Which incidents shaped industrial control security?
  7. Why are Modbus, DNP3 and IEC 104 insecure?
  8. How exposed are HMIs and PLCs on the internet?
  9. How do default and shared passwords lead to incidents?
  10. What happened at Oldsmar, and what does it teach about remote access?
  11. How does ransomware on IT disrupt operations?
  12. What did Stuxnet show about PLC logic?
  13. How did Triton target safety systems?
  14. Which PLC vulnerabilities have major vendors disclosed?
  15. Why are engineering workstations and project files high-value targets?
  16. How should you weigh these risks?
  17. How do you detect attacks on SCADA systems?
  18. What do IEC 62443, the ISA standards and NERC CIP cover?
  19. How should utilities and operators harden SCADA systems?
  20. How should ICS makers build secure products?
  21. Why is ICS security a software supply chain problem?
  22. How Safeguard helps ICS makers and utilities
  23. The takeaway
  24. Frequently asked questions
  25. Sources

What are SCADA, DCS, PLCs, RTUs and HMIs?

SCADA (supervisory control and data acquisition) is the layer of computers and screens that supervises a process. PLCs and RTUs are the rugged controllers that actually switch pumps, valves and gates. HMIs are the operator screens, and a DCS is a plant-wide control system with the same parts packaged by one vendor. Together they are called industrial control systems (ICS).

The vocabulary matters, because attacks land on different layers and each layer fails in its own way.

Industrial control systems (ICS) is the umbrella term for the computers that run physical processes: water treatment, power distribution, pipelines, factories, dams. People also use operational technology (OT) for the same world, to contrast it with information technology (IT), the email, finance and office systems of the same organisation.

SCADA is, in Wikipedia's definition, "a control system architecture comprising computers, networked data communications and graphical user interfaces for high-level supervision of machines and processes." Operators use it to watch the process and issue commands such as setpoint changes, while "the real-time control logic or controller calculations, are performed by networked modules connected to the field sensors and actuators." SCADA was designed for remote access to many sites, and it can span large distances: a water utility might supervise dozens of pump stations, reservoirs and tanks from one control room.

A programmable logic controller (PLC) is "an industrial computer that has been ruggedized and adapted for the control of manufacturing processes," per Wikipedia. PLCs replaced hard-wired relay logic; the first, the Modicon 084, dates to 1968. A PLC reads inputs (a level sensor, a pressure switch), runs a program in a fixed cycle and drives outputs (a pump starter, a valve actuator). It is a hard real-time system, so its outputs must follow its inputs within a set time.

A remote terminal unit (RTU) is a controller built for remote sites. Wikipedia describes it as a device that "interfaces objects in the physical world to a distributed control system or supervisory control and data acquisition (SCADA) system by transmitting telemetry data to a master system, and by using messages from the master supervisory system to control connected objects." RTUs often run on small solar systems and talk over radio, cellular or satellite links.

A human machine interface (HMI) is the operator's screen: a mimic diagram of the plant with live values, alarms and buttons. HMIs range from a panel bolted to a PLC cabinet to a full SCADA workstation, and increasingly to a web page.

A distributed control system (DCS) spreads control across many controllers near the equipment with central operator stations, and first emerged in large, high-value, safety-critical process industries where one supplier delivered the whole stack. Wikipedia notes that today SCADA and DCS functionality "are very similar."

Two more parts appear throughout this guide. The engineering workstation is the PC (often a laptop) that runs the vendor's programming software to write, download and upload controller logic. The safety instrumented system (SIS) is a separate set of controllers whose only job is to bring the process to a safe state when something goes wrong. Wikipedia's article on SIS notes that such systems "must be independent from all other control systems that control the same equipment."

How is a water treatment plant or a dam controlled?

A water plant runs a chain of steps from intake to distribution, each driven by PLCs that dose chemicals and run pumps, supervised by SCADA and operators. A dam uses level and flow sensors, gate hoists and valves run by a local controller, linked by telemetry to a control room, with manual control as the fallback.

Most utilities run both treatment plants and a network of remote sites, so it is worth seeing what software touches.

Layered diagram of a water treatment plant and a dam: supervisory SCADA servers, HMIs, historian and engineering workstation at the top; PLCs and remote RTUs in the middle; the physical treatment process from intake to distribution below; and a dam panel with sensors, spillway gate hoists, minimum flow valve, local controller, telemetry and manual control

Figure 1. Computers supervise, controllers act, and the physical process sits at the bottom.

The water treatment chain

Water purification removes particles, microbes and dissolved substances so water is fit to drink. A conventional surface water plant works roughly like this, and each step has instruments and controllers:

  • Intake. Raw water pumps lift water from a river, lake or reservoir through screens. A PLC starts and stops pumps against demand and tank levels.
  • Coagulation and flocculation. Coagulants such as aluminium sulphate or iron salts are dosed so fine particles clump together. Dosing pumps are controlled by a PLC, usually in proportion to flow and water quality readings.
  • Sedimentation and filtration. The clumps settle or are floated off, and filters remove what remains. PLCs run valves and backwash sequences based on turbidity and pressure.
  • Disinfection and pH control. Chlorine or another disinfectant is dosed, and the pH is adjusted. Wikipedia notes that "lime, soda ash, or caustic soda (sodium hydroxide) can be added to raise the pH," which helps treatment and reduces lead dissolving from old pipes. That chemical is central to the Oldsmar story later in this guide.
  • Clearwell and high lift pumps. Treated water is held while disinfection works, then pumped into the distribution network.
  • Distribution. Tanks, reservoirs, booster pump stations and valves keep pressure across the service area. These are often unstaffed sites run by RTUs or small PLCs over radio or cellular telemetry.

The operator watches all of this on SCADA screens, sees alarms when a value leaves its band, and changes setpoints. The historian stores every reading, which is invaluable for investigations. The engineering workstation holds the PLC programs and is the only machine that should ever change them.

How a dam is controlled

A spillway provides "the controlled release of water from a dam or levee," and spillways "can include floodgates and fuse plugs to regulate water flow and reservoir level," so operators can release water in a controlled way before a reservoir fills. Routine releases for water supply, irrigation or power usually go through outlet works and valves rather than over the spillway.

The control side of a dam is simpler than a treatment plant but just as dependent on software. Level gauges and flow meters feed a local PLC or RTU. That controller runs gate hoists and outlet valves, including the valve that keeps a minimum flow in the river below. Telemetry links the site to a control room that may supervise many dams and reservoirs, sometimes through a web-based HMI. And almost every dam keeps local manual control: a panel at the site, handwheels or local operation of the hoists, and staff trained to use them.

The safety layers that do not depend on software

Calm is warranted here, and the reason is design. Water and dam systems were engineered to be safe when components fail. Treatment plants run independent analysers that check residual chlorine and pH, alarms that wake operators, and limits on dosing pumps. Dams have spillways sized for floods and gates that can be run by hand. Process plants with dangerous chemistry add a separate safety instrumented system, and below that the physical design itself. Mandiant, writing about safety systems, made the same point: if both the safety system and the control system fail, the last layer of protection is the design of the facility itself, including "mechanical protections on equipment (e.g. rupture discs), physical alarms, emergency response procedures and other mechanisms."

In several of the incidents below, those layers did their job. Plants switched to manual, operators reversed changes and a safety system tripped a plant to a safe state. The goal of OT security is to keep it that way: to make sure an attacker cannot remove the margin those layers provide.

What is the Purdue model?

The Purdue model is a reference architecture from the 1990s that splits an industrial organisation into levels, from the physical process (level 0) through controllers, supervision and site operations up to business systems. Security teams use it to place firewalls between levels, especially an industrial DMZ between business IT and plant operations.

The Purdue Enterprise Reference Architecture was developed in the 1990s by Theodore J. Williams and members of the Industry-Purdue University Consortium for Computer Integrated Manufacturing. Its levels, as Wikipedia summarises them:

  • Level 0 "defines the actual physical processes": sensors, pumps, valves, gates.
  • Level 1, "sensing and manipulating the physical processes": PLCs, RTUs and safety controllers.
  • Level 2, "supervising, monitoring and controlling the physical processes": HMIs and operator stations.
  • Level 3, "managing production work flow": SCADA servers, historians, engineering workstations and site services.
  • Level 4, "managing the business-related activities": planning, logistics, billing.

Security practice added two more. Fortinet's description includes Level 5, the enterprise network, and a Level 3.5 DMZ that "acts as a secure buffer between the ICS network and external networks." It recommends that "vendor, contractor, and employee sessions must land on jump servers at Level 3.5" and that DMZ hosts never initiate connections into OT. The related ISA-95 standard from the International Society of Automation (ISA) defines the interface between enterprise and control systems along the same lines.

The Purdue model drawn as seven stacked bands from Level 5 enterprise network to Level 0 physical process, with a Level 3.5 industrial DMZ, and on the right the matching IEC 62443 zones joined by conduits

Figure 2. The Purdue levels, with the IEC 62443 zones and conduits you would draw over them.

Is the model still relevant when sensors send data straight to the cloud? Zscaler notes that the question is often asked, and that many practitioners still think it is too early to discard it. In our view the levels remain the clearest way to explain who should talk to whom. Cloud and edge connections do not remove the need for boundaries; they add new conduits that must be designed as carefully as the old ones.

IEC 62443, covered later, turns this picture into requirements. You group assets into zones by risk and function, and you control every communication path between zones as a conduit. In most plants, the zones line up with Purdue levels, plus a separate zone for the safety system.

What is the attack surface of a SCADA system?

The attack surface is every path to the plant control network: internet-exposed HMIs and controllers, remote access tools, the IT network and its accounts, default passwords, engineering workstations and project files, removable media, integrators and vendors, software updates, telemetry links and physical access to unstaffed sites.

Group the entry points by where the attacker starts, because that decides how many sites one weakness can reach.

From the internet. HMIs, PLCs and gateways that answer on a public address; remote desktop and desktop sharing tools; vendor remote support tunnels and cloud monitoring portals. A weakness here can be found by anyone running a scan.

Through IT and people. Ransomware and intrusions that begin with phishing on the office network and move over shared accounts and file shares; reused passwords; default credentials that were never changed at commissioning.

Through engineering and supply. Engineering workstations and the project files they open; USB drives and visiting laptops; system integrators and maintenance contractors who connect to many sites; software installers, firmware and updates from vendors; radio and cellular links to remote sites; and physical access to pump stations and dam control huts that nobody staffs.

Map of twelve entry points into a SCADA system grouped as from the internet, through IT and people, and through engineering and supply, all leading to the plant control network

Figure 3. Twelve ways in, grouped by where the attacker starts.

The public record has a clear shape. The incidents that reached real utilities in recent years were mostly cheap: an exposed screen, a default password, a remote access tool, ransomware on the office side. The attacks that changed what controllers do were rare, expensive and slow to build. Both matter, but in a different order.

What does an attacker need to attack a SCADA system?

For opportunistic attacks, very little: an internet scan, a default or weak password, and a web browser or the vendor's own software. For attacks that change controller logic or safety systems, a great deal: access to the engineering environment, the vendor's software and hardware, knowledge of the process, and months of preparation.

Here is what each class of attack has needed in the documented cases.

  • Exposed HMI or controller. A search of internet scan data, the device's address, and the default or a weak password. The 2025 Risevatnet dam incident and the 2023 Unitronics incidents fit this profile.
  • Remote access tool abuse. A valid account or shared password for desktop sharing software or a VPN, often reused or leaked.
  • Ransomware reaching operations. A foothold on the IT network, usually from phishing or a stolen credential, and a path to systems the plant depends on, or simply enough disruption that the operator shuts operations down to be safe.
  • Protocol abuse inside the plant network. Network access to the control segment and knowledge of the protocol. Because Modbus, DNP3 and IEC 104 lack built-in authentication, no password is needed once an attacker can send packets.
  • Logic changes and safety system attacks. Remote access to an engineering workstation, a copy of the vendor's software and often the hardware to test against, reverse engineering of proprietary protocols, and deep knowledge of the process. Mandiant concluded that the Triton attacker "must have performed OT reconnaissance and have sufficient specialized engineering knowledge to understand the industrial process being controlled."

That gap in effort explains the risk ranking later in this guide.

Which incidents shaped industrial control security?

A former contractor's sewage release in 2000, Stuxnet in 2010, poisoned vendor installers in 2014, Industroyer in 2016 and Triton in 2017 defined the targeted end. Since 2019, ransomware forcing manual operations and simple attacks on exposed, default-password devices at water utilities and a dam have defined the everyday end.

Vertical timeline of fifteen ICS incidents and research milestones from 2000 to 2025, colour coded as incidents at operators, ransomware reaching operations, and research or measurement

Figure 5. Twenty five years of incidents and research covered in this guide.

The earliest widely cited case is from 2000 in Maroochy Shire, Queensland, where, per Wikipedia's summary, "a former contractor used a stolen laptop and radio equipment to access the sewage control system, releasing millions of liters of raw sewage into local parks, rivers, and the grounds of a hotel." The lessons from that case (insiders, radio telemetry without authentication, slow detection) still appear in incidents a quarter of a century later.

The rest of this guide takes these cases one class at a time.

Why are Modbus, DNP3 and IEC 104 insecure?

Modbus, DNP3 and IEC 60870-5-104 were designed for closed networks, so in their original forms they carry no built-in authentication. Any device that can reach a controller on the network can read values and send commands that look exactly like the operator's. Secure versions exist but are unevenly deployed.

Modbus

Modbus was "originally published in 1979 by Modicon" and "originally designed for use with programmable logic controllers." It runs over serial lines and over TCP, where Modbus TCP uses port 502. It is simple, open and everywhere: water plants, building systems, manufacturing, energy. Wikipedia is blunt about the weakness: "Modbus protocol itself provides no security against unauthorized commands or interception of data." A controller will act on a well-formed write request from any source it can hear. Censys, describing the protocols it found exposed online, put it the same way: Modbus is "widely used across industries, but often lacks encryption and authentication." (Our explainer on Modbus RTU, Modbus TCP and Modbus security goes deeper.)

DNP3

DNP3 was developed in 1993 when GE-Harris Canada built on the then-unfinished IEC 60870-5 specifications for North American utility needs. "Its main use is in utilities such as electric and water companies." Its original design had no authentication, and Wikipedia notes that "much work has been done to add Secure Authentication features," because utility networks are often reachable by third parties. The 2012 IEEE standard (IEEE 1815-2012) "includes features for Secure Authentication Version 5," which can use public key infrastructure and remote key changes, and DNP3 can run over TLS in line with IEC 62351-3. Deployment of those options remains uneven, because both ends must support them and keys must be managed for years. See our DNP3 and Secure Authentication explainer.

IEC 60870-5-104

IEC 60870-5-104 extends the serial IEC 101 protocol over TCP/IP, and the list of registered ports gives 2404 as the port "used to send electric power telecontrol messages between two systems." Wikipedia states that "the security of IEC 104, by design has been proven to be problematic." The IEC committee published IEC 62351 to add encryption and authentication, but "due to the increase in complexity and cost, system owners are reluctant to roll this out on their networks."

Table comparing Modbus, DNP3 and IEC 60870-5-104 by origin and transport, where each is used, built-in authentication and the secure option available

Figure 4. Three common protocols, none with built-in authentication in its original form.

What protocol abuse looks like: Industroyer

The clearest documented demonstration of protocol abuse is Industroyer, analysed by ESET in 2017 after an outage at an electricity substation in December 2016. ESET explained that the malware had four payload components targeting IEC 60870-5-101, IEC 60870-5-104, IEC 61850 and OPC DA, and was "capable of controlling electricity substation switches and circuit breakers directly." The key sentence for defenders: "Industroyer's dangerousness lies in the fact that it uses protocols in the way they were designed to be used." The attackers "didn't need to be looking for protocol vulnerabilities; all they needed was to teach the malware 'to speak' those protocols."

That is the whole problem in one line. When a protocol trusts network position, the network is the only gate.

How to detect it. Passive monitoring of OT traffic can learn which hosts normally send write commands, to which controllers, at what rate. A write from a new source, a burst of commands, or commands outside maintenance windows are strong signals. Many controllers also log mode changes and program downloads.

How to defend. Keep these protocols inside their zone and never route them across the internet. Put conduits (firewalls or gateways that understand the protocol) between the control zone and everything else, and allow only the specific masters that need to talk to each controller. Where devices support them, enable DNP3 Secure Authentication or IEC 62351. For new designs, prefer protocols with security built in, such as OPC UA with signing and encryption (see our OPC UA security guide), and wrap legacy links in authenticated tunnels.

How exposed are HMIs and PLCs on the internet?

Very. Internet scanning research by Censys counted more than 145,000 exposed industrial control system services worldwide in 2024, and over 7,700 exposed HMIs in 80 countries. Many HMIs were reachable with weak or no authentication, and more than a third of one common HMI type belonged to water and wastewater systems.

Internet scanners such as Censys and Shodan continuously probe the IPv4 address space and record what answers. Researchers use the data to measure exposure, and so do attackers. In its 2024 State of the Internet report, Censys said it "identified over 145,00[0] exposed Industrial Control System (ICS) services globally, more than one-third of which are located in the United States," and observed over 148,000 ICS services across 175 countries, with North America hosting 38 percent and Europe 35 percent.

The HMI numbers matter most for utilities. Censys "identified over 7,700 exposed HMIs across 80 countries," and called HMIs "the most concerning and compelling exposures in the ICS space," because "many are accessible without authentication or rely on weak default configurations." Looking at the most common HMI type it found, Censys reported that "more than one-third of exposures were water and wastewater related." It also cited GreyNoise research showing internet-connected HMIs "were scanned and probed more quickly than baseline sensors."

Vendors have said the same thing in plain terms. In May 2024, Rockwell Automation told customers to take devices off the public internet. As reported by The Hacker News, the company said: "Users should never configure their assets to be directly connected to the public-facing internet," and that "removing that connectivity as a proactive step reduces attack surface." The advisory listed known vulnerabilities to patch, including CVE-2021-22681 and CVE-2023-3595, both covered below.

Why it happens. Remote monitoring saves a truck roll to a remote site. A cellular router with a public address, a port forward set up by an integrator, or a web HMI enabled for convenience can expose a controller without anyone deciding to "put it on the internet." Small utilities often lack the staff to notice.

How to detect it. Find out what the internet sees. Your own external attack surface inventory, your ISP and the cellular provider for remote sites can tell you which of your addresses answer on ICS ports or serve HMI pages. Several sector organisations and researchers notify utilities of exposures, and those notices deserve fast action.

How to defend. Remove direct exposure. Every remote path should go through a broker in the DMZ with multi-factor authentication (MFA), not a port forward. Use private APNs or VPNs for cellular telemetry. Disable web servers and remote protocols on devices that do not need them.

How do default and shared passwords lead to incidents?

Many controllers and HMIs ship with a default password, and many sites never change it. When such a device is reachable from the internet, anyone who knows the default can take control. Documented cases include water utilities using Unitronics controllers in 2023 and a Norwegian dam in 2025.

Unitronics controllers at water utilities (2023)

In late November 2023, the Municipal Water Authority of Aliquippa in Pennsylvania reported that a booster station had been compromised. Claroty's Team82 described the device as a Unitronics V570 PLC with a built-in HMI, which was at a minimum defaced. The booster station "did trigger an alarm during the attack to officials who said they immediately shut down the station and began manual operations." The authority's chairman said the pump "was on its own computer network, separated from our primary network and is physically miles away," and officials said public safety was never at risk.

Unitronics then published CVE-2023-6448: VisiLogic before version 9.9.00, "used in Vision and Samba PLCs and HMIs, uses a default administrative password. An unauthenticated attacker with network access can take administrative control of a vulnerable system." Recommended steps reported by Claroty include changing the default password (1111), using MFA for remote access, putting remote connections behind a firewall or VPN, backing up PLC and HMI applications, and changing default ports. Claroty also noted that a Shodan search showed close to 2,000 Unitronics devices facing the internet. Our post on Unitronics default passwords at water utilities has a full checklist.

The Risevatnet dam in Bremanger, Norway (2025)

In April 2025, attackers opened a valve at a dam that serves a fish farm. Claroty reported that "attackers compromised a weak password and were able to manipulate valves at the Lake Risevatnet dam, opening them all the way, causing an increase in water discharge of 497 liters per second above the mandated minimum flow." The entry point "was a web-accessible control panel that managed the valve responsible for the dam's minimum water flow," and the change "went undetected for four hours but did not cause physical damage or endanger public safety."

Radiflow added that the extra outflow "was well within the dam's safe design," noting that the riverbed can handle up to 20,000 litres per second. And WaterWorld reported a detail every dam operator should read twice: the change "was only discovered when a security guard, with no cyber background, noticed an unusual flow and physically checked the valve."

That case is the dam version of the Unitronics story. No vulnerability research, no malware. A password and a web page.

What the attacker needs. The address of an exposed device and a default or guessable password.

Impact. Changes to setpoints, pumps or valves within whatever range the controller allows; defacement; loss of view. In the documented cases, physical design limits and manual intervention kept the outcome safe.

How to detect it. Alarms on setpoint and valve position changes, and on logins to HMIs and controllers, routed to someone who is awake. Compare commanded positions with independent measurements (a flow meter downstream, a level sensor), which is exactly how the Bremanger change was eventually noticed.

How to defend. Change every default password at commissioning and record it in a password manager, not on the panel. Use unique credentials per site. Turn on the controller's own access control features. And never expose an HMI or controller directly to the internet, with or without a strong password.

What happened at Oldsmar, and what does it teach about remote access?

In February 2021, the city of Oldsmar, Florida, reported that someone had remotely raised the sodium hydroxide level at its water plant from 100 to about 11,000 parts per million, and an operator reversed it. In 2023 a former city official said it had been an employee error, and investigators said they could not confirm an outside intrusion.

The Oldsmar story is told in two halves, and both are useful.

What was reported in 2021

According to CyberScoop's reconstruction, on the morning of 5 February 2021 an employee saw someone remotely log in to the system around 8 a.m. and dismissed it, because remote access by colleagues was normal. Around 1:30 p.m., an employee watched as someone with access to the HMI raised sodium hydroxide, the chemical used to control acidity, "from 100 parts per million to 11,000 ppm," and changed it back. Officials said at the time that "multiple safety systems would have prevented any real harm."

Contemporary reporting focused on remote access hygiene. TechTarget reported, citing a state advisory, that "several computers in the Oldsmar plant had TeamViewer installed and all of those computers shared the same password for remote access," that the computers were connected to the internet without firewall protection, and that "all computers used by water plant personnel were connected to the SCADA system and used the 32-bit version of the Windows 7 operating system." CyberScoop notes that TeamViewer as the initial access point "was never confirmed."

The later doubts

In 2023, the city's former manager said at a public administration conference that the incident was a "non-event" caused by an employee, and investigators told CyberScoop that they "were not able to confirm that this incident was initiated by a targeted cyber intrusion of Oldsmar." The Water Information Sharing and Analysis Center had raised the possibility early, asking whether "a level from '100' to '110' or '111' is a plausible change and the trailing zeros were erroneously not deleted." The local sheriff's office said the case remained open.

So the honest summary is this: Oldsmar may not have been an attack at all. It is still one of the most useful cases in water security, for two reasons. First, the conditions described in 2021 (shared remote access credentials, unsupported Windows, no firewall between the internet and machines connected to SCADA) are common and dangerous whatever happened that day. Second, whether the change came from an intruder or a slip of a finger, the protections that matter are the same: dosing limits in the PLC that reject impossible setpoints, a second confirmation for large changes, alarms, and independent analysers downstream. Our post on Oldsmar and remote access lessons and our explainer on why OT remote access is the weak point go further.

How to defend remote access in practice.

  • Remove desktop sharing tools from machines that can control the process. If remote support is needed, use a broker in the DMZ.
  • Require MFA and named accounts for every remote session. No shared passwords.
  • Make vendor sessions time limited and approved per session, and record them.
  • Give remote users view-only access by default; require an on-site confirmation for control actions on chemical dosing and gate positions.
  • Put hard limits in controller logic: a dosing pump should never accept a setpoint far outside the plant's design range, whoever types it.

How does ransomware on IT disrupt operations?

Ransomware usually encrypts office systems, not controllers. But because plants depend on IT for billing, scheduling, accounts and remote access, and because operators cannot always be sure the infection has not crossed over, companies often stop operations or switch to manual as a precaution. Several water utilities, an aluminium producer and a fuel pipeline have done exactly that.

Diagram of ransomware moving from the IT network through shared links such as domain accounts, remote access, file shares and historian connections toward the operations network, with five documented outcomes and a list of what keeps operations running

Figure 8. Ransomware rarely encrypts PLCs; it forces a shutdown or a switch to manual.

The documented cases show a consistent pattern.

Norsk Hydro (2019). One of the world's largest aluminium companies was hit by LockerGoga ransomware. Microsoft's account describes how "production lines had stopped at some of its 170 plants. Other facilities were switching from computer to manual operations." The attack began with an infected email attachment months earlier and eventually cost about 71 million dollars. Hydro refused to pay, restored systems and briefed the public daily. It is the clearest example of a manual fallback keeping a business running.

Colonial Pipeline (2021). On 7 May 2021, the fuel pipeline operator was hit by ransomware and, per Wikipedia's summary of the reporting, "halted all pipeline operations to contain the attack." The same summary notes that the attack "primarily targeted the company's billing infrastructure" and that the company shut the pipeline down "as a precaution." Operations restarted on 12 May. The shutdown was a containment decision, which is exactly why IT ransomware is an operational risk, and the article also records a regulator's view that the company had probably not prepared adequately for shutting down and restarting the pipeline manually.

South Staffordshire Water (2022). The UK supplier, which delivers about 330 million litres of drinking water a day to 1.6 million people, confirmed IT disruption from a criminal attack. BleepingComputer reported the company's statement that "the safety and water distribution systems are still operational," so the IT disruption did not affect the supply of safe water.

Arkansas City, Kansas (2024). The city detected a cyber incident on a Sunday in September 2024. Security Affairs reported the city manager's statement that "out of caution, the Water Treatment Facility has switched to manual operations while the situation is being resolved," and that the water supply "remains completely safe."

American Water (2024). The US water and wastewater utility disclosed unauthorised activity in October 2024. TechTarget reported that the company "proactively took its customer portal offline and has temporarily paused customer billing," while saying "none of its water or wastewater facilities or operations have been negatively impacted." See our post on the American Water incident and utility supply chain lessons.

What the attacker needs. A foothold on the IT network (a phishing email, a stolen VPN password, an unpatched edge device) and time to spread to servers.

Impact. Business disruption, billing and customer service outages, and precautionary shutdowns or manual operation of plants. In all five cases above, the physical process stayed safe.

How to detect it. The usual IT signals (new administrator accounts, mass file changes, disabled security tools, unusual remote access) plus OT-specific ones: authentication attempts from IT into OT, new connections through the DMZ, and changes on historians and SCADA servers.

How to defend.

  • Separate identity: OT systems should not trust the corporate domain. A stolen office password should not log in to a SCADA server.
  • Allow only named, one-way flows through the DMZ. Data can leave OT to IT through a historian copy; commands should not flow back.
  • Keep offline copies of PLC programs, HMI projects, SCADA configurations and network device configs, and test restores.
  • Write, rehearse and time manual operating procedures. Norsk Hydro and Arkansas City could switch to manual because they could.
  • Decide in advance who has the authority to isolate OT from IT, and on what signals.

What did Stuxnet show about PLC logic?

Stuxnet, found in 2010, was the first public case of malware that reprogrammed PLCs. It spread through Windows and USB drives, infected the Siemens engineering software, wrote its own code to specific controllers driving motors, varied their speed outside normal limits, and replayed normal values to operators so the change stayed hidden.

This section describes Stuxnet technically, from the published analysis, and makes no attribution.

Symantec's W32.Stuxnet Dossier summarised its goal plainly: "to reprogram industrial control systems (ICS) by modifying code on programmable logic controllers (PLCs) to make them work in a manner the attacker intended and to hide those changes from the operator of the equipment." The worm was first uncovered in June 2010, and the dossier lists what it combined:

  • Spreading on Windows. It "self-replicates through removable drives exploiting a vulnerability allowing auto-execution" (the Windows shortcut flaw tracked as CVE-2010-2568), spread through a print spooler flaw and an older server service flaw, and exploited "a total of four unpatched Microsoft vulnerabilities." Its drivers were signed with certificates stolen from two hardware companies, which Verisign later revoked.
  • Infecting the engineering environment. It "copies itself into Step 7 projects in such a way that it automatically executes when the Step 7 project is loaded," and infected machines running the WinCC SCADA database using a hard-coded password (CVE-2010-2772).
  • Hijacking the PLC library. Step 7 talks to PLCs through a library called s7otbxdx.dll. Stuxnet renamed the original and replaced it with its own, which forwarded most calls but intercepted the ones that "read, write, and enumerate code blocks on the PLC." That let it "modify the data sent to or returned from the PLC without the operator of the PLC realizing it."
  • Fingerprinting the target. It acted only on specific CPUs (the 6ES7-315-2 in the S7-300 family and the 6ES7-417 in the S7-400 family) with a particular arrangement of frequency converter drives.
  • Changing the physics. On a matching system, it expected drives to run between 807 Hz and 1,210 Hz, and then changed the speed to 1,410 Hz, down to 2 Hz, and back to 1,064 Hz, repeatedly.
  • Hiding the change. It included what the dossier calls "the first ever PLC rootkit," and Wikipedia's summary describes it "returning a loop of normal operation system values back to the users."

Two column diagram: the six steps Stuxnet used to change PLC logic and hide it, and the six steps of the Triton safety system attack, with the lessons both cases teach

Figure 7. Both cases went through the engineering software that programs controllers.

Stuxnet mattered for one reason above all: it proved that controller logic can be changed by software that arrives on the engineering side, and that operators can be shown a false picture while it happens. Every defence below follows from that. Our post on Stuxnet and PLC sabotage lessons for industrial safety covers the engineering side in more depth.

How to detect logic changes. Keep a known good copy of every controller program offline and compare it with what runs on the controller, on a schedule and after any maintenance. Alert on program downloads and controller mode changes. Check process values against independent instruments, because a manipulated HMI can lie but a separate flow meter on a different system is harder to fool. See how to back up PLC logic and check it against a known good baseline.

How to defend. Control removable media in the plant, harden and patch engineering workstations, use controllers' protection features (passwords on program changes, key switches in RUN), and keep engineering software away from email and web browsing.

How did Triton target safety systems?

Triton (also called Trisis) was found in 2017 at a petrochemical plant. It was built to reprogram Schneider Electric Triconex safety controllers. It needed the safety engineering workstation and controllers left in PROGRAM mode, and it exploited a firmware flaw. A failed validation check tripped the plant to a safe shutdown, which led to its discovery.

This section is technical and makes no attribution.

A safety instrumented system watches the process independently of normal control. Mandiant explained that if the process "exceeds the parameters that define a hazardous state, the SIS attempts to bring the process back into a safe state or automatically performs a safe shutdown." Triton was "an attack framework built to interact with Triconex Safety Instrumented System (SIS) controllers."

From Mandiant's analysis:

  • Access. "The attacker gained remote access to an SIS engineering workstation and deployed the TRITON attack framework to reprogram the SIS controllers."
  • Disguise. The malware "was named to masquerade as the legitimate Triconex Trilog application," a log viewer in the TriStation suite.
  • Protocol. It implemented "the proprietary TriStation protocol which is not publicly documented," the same protocol the legitimate software uses to configure controllers.
  • Outcome. "The SIS controllers initiated a safe shutdown when application code between redundant processing units failed a validation check." Mandiant assessed that the attacker "inadvertently shutdown operations while developing the ability to cause physical damage." Wikipedia notes a first shutdown in June 2017 and a second in August 2017 that led to the investigation.

Two details turn this from a story into a checklist. SecurityWeek reported that Schneider Electric confirmed Triton "did in fact exploit a flaw in older versions of the Triconex Tricon system," later published as CVE-2018-7522: in Tricon MP model 3008 firmware 10.0 to 10.4, a flaw "could allow attackers to gain supervisor-level access and control system states." And the malware "can only compromise a SIS device if it's set to PROGRAM mode," which is controlled by a physical key. Mandiant's advice was direct: "On Triconex controllers, keys should not be left in the PROGRAM mode other than during scheduled programming events."

The calm lesson from Triton is that the safety system worked. It detected something wrong and shut the plant down safely. The serious lesson is that someone had spent a long time trying to make it not work. Our post on Triton and the safety system attack and our explainer on safety instrumented systems versus process control systems go further.

How to detect it. Alarm on any key switch change and any program download to a safety controller. Investigate every unexplained trip; the first Triton shutdown was initially treated as mechanical. Monitor traffic to the safety network, which should be almost silent outside planned work.

How to defend. Keep the safety system in its own zone with its own engineering workstation, which should not be reachable from the business network. Keep key switches in RUN and lock them, with changes logged. Patch safety controller firmware during planned outages. And keep mechanical protections, which do not run software.

Which PLC vulnerabilities have major vendors disclosed?

Every major controller vendor has published serious vulnerabilities: authentication bypasses that let an unauthenticated attacker talk to Rockwell and Schneider controllers as if they were the engineering software, a Siemens memory protection bypass, flaws that let program code on the controller differ from what the engineer sees, and boot integrity gaps. Most need network access to the controller.

The table lists documented cases, with the descriptions taken from the CVE records.

CVEVendor and productWhat the record says
CVE-2021-22681Rockwell Automation Studio 5000 Logix Designer and RSLogix 5000 with many Logix controllersAn unauthenticated attacker "could bypass this verification mechanism and authenticate" with CompactLogix, ControlLogix, GuardLogix and other controllers
CVE-2022-1161Rockwell ControlLogix, CompactLogix and GuardLogixLogix Designer "writes user-readable program code to a separate location than the executed compiled code, allowing an attacker to change one and not the other"
CVE-2022-1159Rockwell Studio 5000 Logix DesignerAn attacker with administrator access on the workstation "could inject controller code undetectable to a user"
CVE-2023-3595Rockwell 1756 EN2 and EN3 ControlLogix communication modulesRemote code execution "with persistence on the target system through maliciously crafted CIP messages"
CVE-2024-6242Rockwell ControlLogix 5580 and related modulesA bypass of the Trusted Slot feature that could let an attacker "execute CIP commands that modify user projects and/or device configuration"
CVE-2020-15782Siemens SIMATIC S7-1200 and S7-1500 families and others"A remote unauthenticated attacker with network access to port 102/tcp could potentially write arbitrary data and code to protected memory areas"
CVE-2022-38773Siemens S7-1500 and related CPUsDevices "do not contain an Immutable Root of Trust in Hardware," so an attacker with physical access "could use this to replace the boot image"
CVE-2021-22779Schneider Electric EcoStruxure Control Expert, Modicon M580 and M340Authentication bypass "by spoofing the Modbus communication between the engineering software and the controller"
CVE-2018-7522Schneider Electric Triconex Tricon MP 3008, firmware 10.0 to 10.4Modifying a fixed memory location "could allow attackers to gain supervisor-level access" (exploited by Triton)
CVE-2023-6448Unitronics VisiLogic for Vision and Samba PLCs and HMIsA default administrative password lets an unauthenticated attacker "take administrative control"
CVE-2022-47379CODESYS Control runtimes, including a build for Beckhoff CX controllersAn authenticated remote attacker may use an out-of-bounds write that "can lead to a denial-of-service condition, memory overwriting, or remote code execution"

Three patterns stand out.

The engineering protocol is the trust boundary. CVE-2021-22681 and CVE-2021-22779 both let an attacker impersonate the engineering software. Once you can do that, you can do what an engineer can do: stop the controller, change logic, change configuration. That is why vendors pair these advisories with advice to segment networks and restrict who can reach controllers. Our post on CVE-2021-22681 has the details.

What the engineer sees may not be what runs. CVE-2022-1161 and CVE-2022-1159 describe the Stuxnet problem in modern products: source and compiled code stored separately, or code injected in a way the user cannot see. That is why comparing logic against an offline baseline, and verifying it with vendor tools built for that purpose, matters.

Boot integrity is a hardware question. CVE-2022-38773 needs physical access, but it is a reminder that a controller is only as trustworthy as its boot chain. Newer hardware generations add roots of trust; older ones will be in plants for decades.

How to defend. Track every controller model and firmware version, subscribe to each vendor's security advisories, and plan firmware updates into outages. Where updates cannot happen quickly, use the mitigations vendors publish (network restrictions, controller protection settings, key switches) and record the risk decision. Our guide to vulnerability management on the factory floor under IEC 62443 covers the process.

Why are engineering workstations and project files high-value targets?

The engineering workstation holds the vendor software, credentials and project files that can change any controller it can reach. Stuxnet infected project files, Triton ran from a safety engineering workstation, and research has shown engineering software that trusts whatever a controller or project folder hands it. Compromise the workstation and you hold the keys to the plant.

Think of the engineering workstation as the plant's master key. It has:

  • the vendor programming software that can stop controllers, download new logic and change configuration;
  • project files containing every controller's program, network addresses and often credentials;
  • network access to every controller it maintains, sometimes across many sites;
  • and, too often, email, a web browser and USB ports.

The documented cases hit each of those.

Project files that run code. Stuxnet copied itself into Step 7 projects so that it ran when the project was opened. A separate Siemens record, CVE-2012-3015, describes how STEP 7 before 5.5 SP1 could load "a Trojan horse DLL in a STEP7 project folder." A project file shared by email or carried on a USB drive is effectively executable content.

Controllers that attack engineers. In 2022, Claroty's Team82 published the Evil PLC attack, with proof-of-concept exploits against engineering software from seven vendors: Rockwell Automation, Schneider Electric, GE, B&R, XINJE, OVARRO and Emerson. A compromised PLC could hand crafted data to the engineering workstation during an upload, and the workstation would run attacker code. Team82 stressed that "all the vulnerabilities we found were on the engineering workstation software side and not in the PLC firmware," and that "in most cases, the vulnerabilities exist because the software fully trusted data coming from the PLC." One of the scenarios it described was "attacking traveling integrators," because an integrator's laptop visits many sites. Most of the affected vendors remediated or mitigated the issues.

Workstations that inject code. CVE-2022-1159 shows that an attacker with administrator rights on a workstation running Studio 5000 could inject controller code the user would not see.

How to detect it. Treat the workstation as a monitored asset: application allowlisting, logging of every connection it makes to controllers, alerts on program downloads outside change windows, and integrity checks on the vendor software's own files (Stuxnet replaced a DLL).

How to defend.

  • Dedicated engineering workstations, kept in the operations zone, with no email or general web browsing.
  • Separate accounts and MFA for engineering access; no shared logins.
  • Patch the vendor engineering software as carefully as the controllers.
  • Control removable media, and scan anything coming in at a kiosk outside the control zone.
  • Require integrators and contractors to use your workstations, or to meet your hardening standard, before they connect.
  • Keep project files in version control with access logs, and compare what you load with what you approved.

Our post on hardening engineering workstations that program PLCs has a full build standard.

How should you weigh these risks?

Rank by effort and impact. Internet-exposed HMIs with weak passwords, default PLC passwords, misused remote access and ransomware reaching operations are cheap for attackers and have disrupted real utilities, so fix them first. Logic and safety system attacks are rare and expensive, and the controls that stop them overlap with the basics.

Risk matrix plotting ten SCADA attack classes by attacker effort and typical impact, with exposed HMIs, default passwords, remote access misuse and ransomware in a highlighted fix first area

Figure 6. Cheap attacks with real impact are where most documented incidents happen.

The placement is a judgement from the documented cases, not a statistic. It points to a practical order of work: remove internet exposure and default passwords this month, fix remote access next, then separate IT from OT properly, then build the monitoring and engineering controls that catch the rarer, harder attacks. The good news is that the later controls (segmentation, logic baselines, workstation hardening) also make the cheap attacks harder.

How do you detect attacks on SCADA systems?

Start with an accurate asset inventory, then watch three things: who connects to controllers and with what commands, whether controller logic and configuration still match a known good baseline, and whether process values agree with independent measurements. Route alarms to people who can act, and investigate unexplained trips.

Signals for small utilities and plant operators

  • Unexpected setpoint changes, valve or gate movements, or pump starts, especially outside staffed hours.
  • Logins to HMIs, SCADA servers or remote access tools from unfamiliar places or at odd times.
  • An HMI that looks different, has been defaced, or shows values that do not match what staff see on site.
  • Downstream or independent instruments that disagree with what SCADA reports (the Bremanger change was noticed this way).
  • Unexplained controller faults or trips. Triton's first shutdown was treated as a mechanical problem.

Signals for larger operators

  • OT network monitoring. Passive sensors on SPAN ports learn normal traffic and alert on new hosts, new protocols, write commands from unexpected sources, program downloads and controller mode changes. This is the core job of OT monitoring platforms.
  • Logic and configuration change detection. Scheduled comparison of controller programs and device configurations with approved versions.
  • Identity and remote access logs. Every session through the DMZ broker, tied to a named person and a change ticket.
  • IT signals that precede OT impact. Ransomware precursors on the IT network, failed authentication from IT into OT, and new connections through the DMZ.

Signals for ICS makers

  • Reports from researchers and customers, through a published disclosure channel.
  • New CVEs in components you ship (TCP/IP stacks, runtimes, open source libraries), matched against what each product version actually contains.
  • Exposure data showing your products on the internet in large numbers, which may mean your defaults or documentation encourage it.

Detection is only useful with response. Write an OT incident response plan that says who can isolate networks, how to switch to manual, how to preserve evidence on controllers and workstations, and how to restore from known good backups. Rehearse it with operators, not just the security team.

What do IEC 62443, the ISA standards and NERC CIP cover?

IEC 62443 (also published by ISA as ISA/IEC 62443) is the main international series for industrial cybersecurity, covering asset owners, service providers and product makers, including zones and conduits, security levels and secure product development. ISA also publishes ISA-95 for enterprise and control integration, and electric utilities in its scope also work to NERC CIP.

ISA/IEC 62443

The ISA Global Cybersecurity Alliance's quick start guide describes the series as addressing "the security of industrial automation and control systems (IACS) throughout their lifecycle," developed by the ISA99 committee and adopted by the IEC. It is written for three audiences: asset owners, product suppliers and service providers. The parts most utilities and makers meet first:

  • Part 2-1: establishing an IACS security programme (asset owners).
  • Part 2-3: patch management in the IACS environment.
  • Part 2-4: security programme requirements for service providers such as integrators and maintenance contractors.
  • Part 3-2: security risk assessment and system design, including zones and conduits.
  • Part 3-3: system security requirements and security levels.
  • Part 4-1: the secure product development lifecycle for product suppliers.
  • Part 4-2: technical security requirements for components such as controllers and HMIs.

Zones and conduits. "A zone is defined as a grouping of logical or physical assets based upon risk or other criteria such as criticality of assets, operational function, physical or logical location, required access or responsible organization." "A conduit is defined as a logical grouping of communication channels that share common security requirements connecting two or more zones." Part 3-2 walks the asset owner through identifying the system, assessing risk, partitioning it into zones and conduits, and documenting requirements.

Security levels. The guide defines security level as "the measure of confidence that the SuC, zone or conduit is free from vulnerabilities and functions in the intended manner," with levels running from protection against "casual or coincidental violation" to protection against attackers "using sophisticated means with extended resources, IACS-specific skills, and high motivation." Requirements are organised under seven foundational requirements, from identification and authentication control to resource availability. Our post on IEC 62443 roles, zones, conduits and safety is a practical introduction, and our IEC 62443-4-1 explainer covers the product side.

Other ISA and IEC standards you will meet

ISA-95 defines the interface between enterprise and control systems and gives the shared vocabulary behind the Purdue-style levels. For safety systems, IEC 61511, published in 2003 and based on IEC 61508, guides the application of safety instrumented systems in the process industries; security risk to the SIS should be part of that assessment.

NERC CIP

Electric utilities in scope of NERC CIP will already know it well; our post on energy sector software security and NERC CIP covers its supply chain requirements.

How should utilities and operators harden SCADA systems?

Remove internet exposure, change every default password, replace ad hoc remote access with a brokered, MFA-protected path, separate IT from OT with zones, conduits and an industrial DMZ, keep offline backups of controller logic, monitor for changes, and rehearse manual operations and incident response.

Eight stacked layers of protection narrowing toward the physical process: governance, asset inventory, segmentation, remote access, device hardening, monitoring, independent safety and recovery

Figure 9. Each layer catches what the one above it missed.

Small utilities and plant operators

You may have one or two people who look after everything. Start here.

  • Take HMIs and PLCs off the internet. Find every device that answers on a public address and remove the exposure.
  • Change every default password on controllers, HMIs, routers and remote access tools, and use unique passwords per site.
  • Replace desktop sharing tools on SCADA machines with one brokered remote access path, with MFA, named accounts and logging.
  • Keep offline copies of PLC programs and HMI projects, labelled with dates, and check you can restore them.
  • Put safe limits in controller logic for chemical dosing and gate positions, so impossible setpoints are rejected.
  • Write and practise manual procedures for each process, and time how long the switch takes.
  • Join a sector information sharing group for alerts and peer advice. In the United States, for example, the Water Information Sharing and Analysis Center serves water and wastewater utilities.
  • Ask your vendors and integrators for SBOMs, patch notes and their remote access practices.

Larger operators and integrators

  • Asset inventory. Every controller, HMI, RTU, server, network device and the software versions on them, kept current. See you cannot secure what you cannot see.
  • Zones and conduits per IEC 62443-3-2, with an industrial DMZ at Level 3.5, a separate zone for safety systems, and written rules for every conduit.
  • No direct IT to OT flows. Data leaves OT through a historian copy or gateway; commands and remote sessions are brokered in the DMZ.
  • Separate OT identity and administrator accounts, with MFA, privileged access management and no trust of the corporate domain.
  • Passive OT network monitoring and logic change detection, with alerts routed to a team that knows the process.
  • Engineering workstation standard: dedicated, hardened, patched, no email or browsing, removable media controlled.
  • Patch and update programme per IEC 62443-2-3, aligned with planned outages, with risk decisions recorded where patches must wait.
  • Rehearsed OT incident response, including isolation decisions, manual operation and controller restores.
  • Supplier requirements in contracts: SBOMs, vulnerability notification, secure remote access, and IEC 62443-2-4 practices for integrators.

Three column hardening checklist for small utilities and plant operators, larger operators and integrators, and ICS makers

Figure 10. Hardening checklists for the people who run and build industrial control systems.

What residents can do

If you are reading this as a customer of a water utility, the reassuring part is that the documented incidents were caught by alarms, staff and design limits, and treated water passes through independent checks. If your utility tells you to boil water or not to drink it, follow that advice, whatever the cause. And if you sit on a town council or a utility board, ask whether the steps in the small utilities list above are done.

How should ICS makers build secure products?

Build to IEC 62443-4-1 and 4-2: no default or hard-coded credentials, authenticated protocols on by default, signed firmware with a hardware root of trust, engineering software that treats controller and project data as untrusted, an SBOM for every release, a disclosure programme and long support windows.

The incidents and CVEs above map almost one to one onto product decisions.

  • Credentials. Ship without default passwords; force a unique password or a provisioning step at first use. CVE-2023-6448 and the Bremanger dam both turned on a password that should never have existed.
  • Authenticated protocols by default. Support DNP3 Secure Authentication, IEC 62351 and OPC UA security, and make the secure mode the default, not an option buried in a manual.
  • Engineering protocol hardening. Authenticate the engineering software to the controller cryptographically. CVE-2021-22681 and CVE-2021-22779 show what happens when that check can be bypassed.
  • Untrusted input in engineering software. Treat data uploaded from a controller and files in a project folder as untrusted. The Evil PLC research and CVE-2012-3015 are the cautionary tales.
  • Logic integrity. Make it possible to verify that the code running on a controller matches the code the engineer approved, and show the user when it does not (CVE-2022-1161).
  • Boot integrity. Signed firmware, verified at boot by an immutable hardware root of trust (CVE-2022-38773).
  • Secure deployment guidance. Documentation that never suggests port forwarding, and features that warn when a device is reachable from a public address.
  • SBOMs and vulnerability handling. Know every component in every firmware and software release, monitor them for new CVEs, publish clear advisories, and fix older product lines that are still in the field.
  • A secure development lifecycle per IEC 62443-4-1, with threat modelling, code review, static analysis and security testing for every release.

Why is ICS security a software supply chain problem?

A controller's firmware contains code from chip vendors, RTOS and network stack suppliers, runtime makers and open source projects. A flaw in one shared component can sit in hundreds of products from many brands at once, and the plant operator usually cannot see it. SBOMs at every link are how that flaw gets traced and fixed.

The ICS software supply chain drawn as seven links from chip SDKs through RTOS and network stacks, control runtimes, device firmware, engineering software and integrator projects to the plant, with documented vulnerabilities under each link

Figure 11. Every controller carries code from many companies it never names.

Follow a single controller back up its supply chain.

Chip and board SDKs. Bootloaders, drivers and security features come from the silicon vendor and board maker. Our post on fabless chip designers and the software supply chain explains how that code travels.

Operating systems and network stacks. Many controllers run a real-time operating system and a licensed TCP/IP stack. In 2021, Forescout and JFrog disclosed INFRA:HALT, 14 vulnerabilities in the NicheStack TCP/IP stack. Help Net Security reported that the stack is "used in OT devices such as the extremely popular Siemens S7 PLCs," that major automation vendors were listed as customers of its original developer, and that "according to an old InterNiche website, the stack is used by almost 200 device vendors." Patches came from the stack's owner, but "device vendors using the stack should provide their own updates to customers." Similarly, CVE-2019-12256 is a stack overflow in the parsing of IPv4 options in Wind River VxWorks 6.9 and vx7, a real-time operating system found in embedded devices.

Control runtimes. Some PLC makers build on a licensed runtime rather than writing their own; the CODESYS records, for example, list a CODESYS runtime built for Beckhoff controllers. In 2023, Microsoft researchers disclosed 16 high-severity vulnerabilities in the CODESYS V3 SDK. Security Affairs reported that they "affect all versions of CODESYS V3 prior to version 3.5.19.0," that exploitation required authentication and deep protocol knowledge, and that users should check with their device manufacturers for patched firmware.

Open source libraries. Gateways, HMIs and edge devices often include open source protocol libraries. CVE-2024-10918, for example, is a stack-based buffer overflow in libmodbus 3.1.10 when replying to a request with an unexpected length.

Device firmware. The vendor's own code, plus everything above, plus the hardware root of trust (or its absence).

Engineering software and integrator projects. Windows applications with their own dependencies, and project files that carry logic and sometimes code.

Distribution. In 2014, F-Secure found that attackers had compromised the websites of three ICS software vendors and replaced their downloads. CSO Online reported that "the software installers available on the sites were trojanized to include the Havex RAT," and that a component scanned local networks for devices answering OPC requests. Two of the vendors made ICS remote management software.

Why SBOMs matter here

A software bill of materials (SBOM) is a machine-readable list of every component and version in a product. When INFRA:HALT or the CODESYS flaws were disclosed, the hardest question for operators was simple: which of my devices contain this? Without SBOMs, the answer came slowly, vendor by vendor. With SBOMs from every maker, a utility can answer it from its own inventory, and a maker can tell which product lines and firmware versions need a fix. IEC 62443-4-1 expects product suppliers to manage third-party components, and buyers increasingly ask for SBOMs in contracts. In Europe, the Cyber Resilience Act adds SBOM obligations for products with digital elements. Our posts on SBOM requirements for industrial control systems and SBOMs for firmware and IoT devices cover formats and practice.

There is an honest limit here. An SBOM does not patch anything, and a plant may not be able to update a controller until the next outage. But it changes the conversation from "are we affected?" to "how do we mitigate until we can update?", which is the conversation a utility needs to have.

How Safeguard helps ICS makers and utilities

Safeguard covers the software supply chain inside industrial products in one platform: SBOMs for firmware, engineering software and apps, binary and image scanning, vulnerabilities ranked by reachability, SCA, SAST and secrets scanning on code you own, fix pull requests gated by your CI, Gold patched packages, supplier risk and compliance evidence, deployed as SaaS, in your VPC or on premises.

Safeguard does not sit on the plant network or watch Modbus traffic. It sits where much of the long-lived exposure in this guide comes from: the firmware, runtimes, engineering tools, HMI software, gateways and open source components that ICS makers build and that utilities buy. Teams that take this seriously often end up with one tool for source scanning, another for binary analysis, another for SBOM management, a questionnaire tool for suppliers and a GRC platform for audits. Safeguard does that work end to end, with one inventory and one vulnerability model, from the first SBOM to the merged fix and the audit record.

Flow diagram of Safeguard for ICS products: source repositories, binaries and images, extracted firmware and supplier SBOMs feed SBOMs, reachability-ranked vulnerabilities, code scanning and policy gates, leading to fix pull requests, Gold patched packages, supplier risk, compliance evidence and SBOM delivery

Figure 12. How Safeguard fits an ICS product or a utility's software estate.

See inside the binaries and images you ship and receive

In the Safeguard console, Integrations > Upload App Binary accepts compiled software: Windows EXE, MSI and DLL files, Android APKs and iOS IPAs, VM and disk images in QCOW2, VMDK, VHD and OVA formats, and container tarballs. Safeguard unpacks the file, identifies embedded libraries and their versions, maps them to known vulnerabilities and records licences, and the result becomes a project with a full SBOM. For ICS teams that covers engineering software and SCADA or HMI installers on Windows, operator and field apps on both mobile platforms, virtual appliances for SCADA servers and historians, and containers that run on edge gateways. Firmware images for controllers and RTUs may need to be extracted into their file systems first, and the usual limits of binary analysis apply: heavy obfuscation, encryption or unusual packing can give incomplete results. Utilities can use the same upload to see what is inside software they receive from vendors before it goes near the plant. Our guide to scanning app binaries with Safeguard and our explainer on binary composition analysis go further.

Rank vulnerabilities by what can actually be reached

A gateway image or an HMI package can contain hundreds of components, and many vulnerabilities in them sit in code the product never calls. Safeguard's reachability analysis ranks the vulnerabilities your code can actually reach above those it cannot, and adds EPSS (the Exploit Prediction Scoring System) and known exploited context. For an industrial product that is how a team separates the protocol parser flaw that needs a fix in the next release from an unused function in a bundled library, and how a utility decides which vendor update to schedule into the next outage.

Scan the code you own

For controller firmware, gateway services, HMI applications, engineering tools and cloud backends that you write yourself, Safeguard runs software composition analysis (SCA) across the full transitive dependency tree, static application security testing (SAST) on your own code, and secrets scanning that catches hard-coded passwords, keys and tokens before they ship. Default and hard-coded credentials (CVE-2023-6448, CVE-2010-2772) and parsers that trust their input (the Evil PLC class) are exactly the classes these scanners exist to catch.

SBOMs your customers and auditors can use

Every project produces an SBOM you can export in CycloneDX or SPDX, among other formats. That is what a utility will ask a controller maker for, what an integrator will ask a gateway supplier for, and what lets a maker answer "which firmware versions ship NicheStack, which CODESYS runtime, which libmodbus" in minutes. Safeguard's Third-Party Risk Manager lets you request SBOMs from your own suppliers, score them and monitor them continuously, so the upstream side of the chain is visible too. See Safeguard TPRM for third-party risk.

Fix it without waiting for upstream

This is where Safeguard differs most from a scanner. For code in your repositories, Griffin, Safeguard's remediation model, opens a fix pull request that your own CI builds and tests before anyone merges it, and your policies decide what can merge. For open source components where upstream has not shipped a fix, or where the fix sits in a version you cannot move to without a larger change, Safeguard Gold provides patched, hardened builds of those packages, and Safeguard can fork and patch the vulnerable package so your next build pulls a fixed version. For an ICS maker with product lines that must stay supported for many years, that is the difference between shipping a fixed firmware in the next maintenance release and carrying a known vulnerability until a major upgrade. Licensed stacks and runtimes still have to be patched by their owners, but the open source around them no longer has to wait.

Compliance evidence from the same data

Safeguard's compliance suite collects evidence for frameworks such as SOC 2 and ISO 27001 from the same scans, policies and fix history. For makers working toward IEC 62443-4-1 or preparing for the EU Cyber Resilience Act, and for utilities answering customer and regulator questions, the SBOMs, vulnerability monitoring records, risk decisions and fix trail are the raw material those assessments ask for.

Deploy where your code lives

Controller firmware, engineering projects and utility software inventories are sensitive. Safeguard runs as SaaS, in your own VPC, or fully on premises. It can run in Zero AI mode with no language model in the loop, or with a model you choose.

How this compares

Industrial teams often combine OT network security platforms such as Claroty, Nozomi Networks and Dragos, which provide OT network visibility and threat detection, with separate tools for source scanning, firmware analysis, SBOM management and audits. The OT platforms do a job that Safeguard does not: watching plant traffic and controllers in real time, which is the core of the detection advice above. Safeguard's case is consolidation and remediation across the software itself: one inventory across source, binaries and images; one vulnerability model with reachability; fix pull requests and patched packages instead of tickets; and supplier risk and compliance evidence from the same data. The two sit well side by side. For more context, see our firmware supply chain security guide and our plan for software supply chain security at water utilities.

The takeaway

Industrial control security has a dramatic reputation, and Stuxnet and Triton earned it. But the incidents that reached water utilities and a dam in recent years were plain. A panel on the internet with a weak password. A controller still using its factory default. Remote access tools with a shared password. Ransomware on the office network that left managers no safe choice but to run the plant by hand.

The protections are just as plain, and they work. Take control systems off the internet. Change every default. Broker remote access. Separate IT from OT with zones and conduits. Keep offline copies of controller logic and compare against them. Keep manual procedures and safety systems that do not trust software. Then do the harder supply chain work: know what is inside every controller, gateway and engineering tool, and be able to fix it without waiting for every supplier upstream. The safety layers in these plants did their job in every case in this guide. The point of security is to keep it that way.

Try it yourself

Click through these interactive Safeguard demos to see the steps described in this guide.

Test yourself

What is the difference between SCADA and a PLC?

SCADA is the layer of computers and screens that supervises a process. PLCs and RTUs are the rugged controllers that actually switch pumps, valves and gates, and HMIs are the operator screens.

Why are Modbus, DNP3 and IEC 104 considered insecure?

They were designed for closed networks, so in their original forms they carry no built-in authentication. Any device that can reach a controller can read values and send commands that look exactly like the operator's.

What did Stuxnet do that made it a turning point?

Stuxnet, found in 2010, was the first public case of malware that reprogrammed PLCs. It spread through Windows and USB drives, wrote its own code to specific controllers, varied motor speeds outside normal limits, and replayed normal values to operators so the change stayed hidden.

What did the Triton malware go after, and how was it discovered?

Triton, found in 2017 at a petrochemical plant, was built to reprogram Schneider Electric Triconex safety controllers. A failed validation check tripped the plant to a safe shutdown, which led to its discovery.

In ransomware cases like Colonial Pipeline and Norsk Hydro, why did operations stop if controllers were not encrypted?

Plants depend on IT for billing, scheduling, accounts and remote access, and operators cannot always be sure the infection has not crossed over. So companies often stop operations or switch to manual as a precaution.

What is a practical defence against default passwords and exposed devices?

Change every default password at commissioning, use unique credentials per site, and never expose an HMI or controller directly to the internet. Every remote path should go through a broker in the DMZ with multi-factor authentication, not a port forward.

Frequently asked questions

What is SCADA, and how is it different from a PLC or a DCS?

SCADA (supervisory control and data acquisition) is the supervisory layer: servers, operator screens and communications that watch a process and send commands, often across many remote sites. A PLC is a rugged industrial computer that directly runs pumps, valves and motors. A DCS is a plant-wide control system with controllers and operator stations from one vendor. In practice SCADA and DCS now overlap, and both rely on PLCs or similar controllers.

Can hackers poison a city's water supply?

It is very hard, and no documented case has done it. Treatment plants run independent analysers, alarms, dosing limits and trained operators, and treated water is checked before and after it leaves the plant. Real incidents have involved exposed controllers with default passwords, remote access misuse and ransomware forcing manual operation. The 2021 Oldsmar case, first reported as an attempt to raise a chemical level, was later described by a former city official as employee error.

Can a dam be hacked?

A dam's control system can be. In April 2025, attackers used a weak password on a web-accessible control panel to fully open a minimum flow valve at the Risevatnet dam in Norway for about four hours. The extra flow stayed within the dam's safe design and caused no damage. Dams have spillways sized for floods and manual control of gates, which is why removing internet exposure and default passwords matters so much.

What did Stuxnet actually do?

Stuxnet, found in 2010, was Windows malware that spread through USB drives and network flaws, infected Siemens Step 7 engineering software and project files, and replaced the library Step 7 uses to talk to PLCs. On specific controllers driving motors through frequency converters, it wrote its own code that changed motor speeds outside normal limits, while replaying normal values to operators. It proved PLC logic can be changed and hidden.

Why are Modbus and DNP3 considered insecure?

Both were designed for closed networks and, in their original forms, have no built-in authentication. Any device that can reach a Modbus or DNP3 controller can send read and write commands that look legitimate. DNP3 later added Secure Authentication version 5 in IEEE 1815-2012, and both can be wrapped in authenticated tunnels, but many installations still rely only on network separation.

What is the Purdue model, and is it still relevant?

The Purdue model is a 1990s reference architecture that splits an industrial organisation into levels: the physical process at level 0, controllers at level 1, supervision at level 2, site operations at level 3 and business systems at level 4, with an industrial DMZ at level 3.5 in security practice. Cloud and edge connections blur it, but it remains the clearest way to decide which systems may talk to which.

What is IEC 62443?

IEC 62443, also published as ISA/IEC 62443, is the main international series of standards for industrial automation and control system security. It covers security programmes for asset owners, requirements for integrators and service providers, risk assessment with zones and conduits, system and component security levels, and the secure development lifecycle for product makers in Part 4-1.

What is an ICS SBOM, and why do utilities ask for one?

An SBOM is a machine-readable list of every software component and version inside a product, such as a PLC's firmware or an HMI package. Utilities ask for them because a flaw in a shared component, like the NicheStack TCP/IP stack used by almost 200 device vendors, can sit in many brands at once. With SBOMs, a utility can find affected devices quickly and plan mitigations.

Sources

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

We email you a link to confirm, and only then add you to the list. Unsubscribe in one click. Privacy

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.