Power grids get hacked through the same front doors as any company, and then through features the grid was built to have. In every publicly confirmed cyber attack that cut power, the attackers first got into business networks through phishing or exposed remote access, spent weeks or months learning the control systems, and then sent commands that operator screens and grid protocols are designed to obey. A growing share of the risk now sits at the edge: smart meters, solar inverters, home batteries and the vendor clouds that manage them. Most of it is software, and most of that software comes from equipment makers and their suppliers.
That is the short answer. This guide is the long one. It explains how the grid works, from power stations through transmission, substations and distribution, and what software runs at each step: energy management systems, SCADA, protection relays, IEC 61850 and GOOSE, smart meters and the distributed energy that now sits on rooftops. It walks through the 2015 and 2016 Ukrainian grid outages as technical events, Industroyer as malware that spoke grid protocols, the idea behind Aurora-style physical damage, the 2023 wave of firewall intrusions at Danish energy companies, and the solar inverter research of 2024 and 2025, including Forescout's SUN:DOWN report. It closes with detection, IEC 62443, IEC 62351 and NERC CIP, hardening checklists for utilities, solar owners and equipment makers, and the software supply chain inside grid equipment.
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, payloads or step-by-step instructions. The tone is calm on purpose. The grid has layers of automatic protection and people who can run it by hand, and both feature in every real case below.
A note on the law. Testing grid equipment, meters, inverters or utility systems that you do not own, or have no written permission to test, is illegal in most countries, and on live grid equipment it can also be dangerous. Everything in this guide is for protecting systems you own or operate, systems you are authorised in writing to assess, or products you build. The researchers cited here worked in their own labs, on their own equipment or through coordinated disclosure.
This post is part of a 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 machines that utility staff use every day, and the guide on how cars and EV chargers get hacked covers the charging side of the grid in depth. Our posts on fabless chip designers and the software supply chain and how a laptop is built explain where the chips and SDKs inside relays, meters and inverters come from.
Contents
- How does the electric grid work, and what software runs it?
- What is the attack surface of the grid?
- What does an attacker need to disrupt the grid?
- What happened in the 2015 Ukrainian grid outage?
- What happened in 2016, and how did Industroyer work?
- Can a cyber attack physically damage grid equipment?
- How are substations and IEC 61850 networks attacked?
- How does vendor remote access put the grid at risk?
- How do smart meters get hacked?
- How do solar inverters, batteries and EV chargers get hacked?
- How should you weigh these risks?
- How do you detect attacks on grid systems?
- What do IEC 62443, IEC 62351 and NERC CIP require?
- Why is grid security a software supply chain problem?
- How should utilities and grid operators secure their systems?
- How should solar, battery and EV charger owners protect themselves?
- How should grid equipment makers build secure products?
- How Safeguard helps teams that build and run grid software
- The takeaway
- Frequently asked questions
- Sources
How does the electric grid work, and what software runs it?
The grid moves power from generators through high voltage transmission lines and substations to local distribution feeders and customers. Software watches and steers every step: energy management systems and SCADA in control centres, remote terminal units and gateways in substations, protection relays that trip breakers in milliseconds, smart meters at homes, and inverters and their clouds at the edge.
An electrical grid is an interconnected network that carries electricity from producers to consumers. It has four physical parts: power stations that generate electricity, substations that step the voltage up or down, transmission lines that carry power long distances at high voltage, and distribution networks that bring it to homes and businesses at lower voltage (Wikipedia, Electrical grid). Large grids are synchronous, which means every generator connected to them turns in step at the same frequency, 50 hertz in Europe and much of Asia and 60 hertz in North America. Keeping supply and demand balanced second by second keeps that frequency steady.
Figure 1. How the grid works. Power flows along the top; control and data flow through software underneath.
Generation
Power stations run their own plant control systems, often a distributed control system (DCS) that manages turbines, boilers, fuel and cooling. Each generator also has protection that disconnects it if something goes wrong. Wind and solar farms add power electronics and plant controllers, and many of them are monitored remotely by the operator or the turbine maker.
Transmission and substations
A substation is the part of the grid where voltage is transformed, lines are switched and faults are isolated. Between a generator and a home, power can pass through several substations at different voltages, and there were around 55,000 substations in the United States alone as of 2022 (Wikipedia, Electrical substation). Inside each one you will find transformers, circuit breakers, disconnect switches and a building full of electronics.
The most important of those electronics are protective relays. A protective relay is a device that detects a fault, such as overcurrent, over or under voltage, reverse power or abnormal frequency, and trips a circuit breaker to isolate it (Wikipedia, Protective relay). Modern relays are microprocessor based. They convert current and voltage measurements into digital form and run protection logic in software, often combining what used to take several electromechanical devices in one box. In today's terminology they are one kind of intelligent electronic device (IED), a catch-all name for networked controllers, relays and meters inside a substation.
Next to the relays sit remote terminal units (RTUs) and gateways. An RTU collects measurements and status from the substation and carries out commands from the control centre, such as opening or closing a breaker. Older substations used serial links, so many still have serial-to-Ethernet converters that translate between the control centre's IP network and serial devices. Those converters became famous in 2015 for a reason we will come to.
Distribution
Distribution substations step voltage down for feeders, the local lines that run along streets. Utilities increasingly automate feeders with remotely controlled switches and reclosers, so a fault can be isolated and power rerouted without a truck roll. A distribution management system (DMS) in the control centre models the distribution network and lets operators switch it remotely.
Control centres: EMS and SCADA
SCADA stands for supervisory control and data acquisition. It is the software that gathers measurements from RTUs and IEDs, shows them to operators on human machine interface (HMI) screens, and sends their commands back out. An energy management system (EMS) is the suite of applications that grid operators use to monitor, control and optimise generation and transmission, and the two are so intertwined that the industry often writes SCADA/EMS or EMS/SCADA (Wikipedia, Energy management system). The control centre servers typically run Windows or Linux, with databases, historians that store time series data, and links to neighbouring utilities.
The protocols
Control centres talk to substations through a handful of protocols, and an attacker who learns them can talk to the grid too.
- DNP3 (Distributed Network Protocol 3) is used mainly by electric and water utilities, between control centres, RTUs and IEDs. Later versions added Secure Authentication, standardised in IEEE 1815-2012, which can use public key infrastructure (Wikipedia, DNP3).
- IEC 60870-5-101 and -104 are the telecontrol protocols common in Europe and elsewhere. 101 runs over serial links and 104 carries the same messages over TCP/IP. IEC TS 60870-5-7 later defined security extensions for both based on IEC 62351 (Wikipedia, IEC 60870-5).
- IEC 61850 is the international standard for communication inside substations. It defines data models for IEDs and maps them onto protocols, including Manufacturing Message Specification (MMS) for client and server traffic, Generic Object Oriented Substation Events (GOOSE) for fast peer messages, and Sampled Values (SV) for streamed measurements (Wikipedia, IEC 61850).
GOOSE deserves its own paragraph. It is how one relay tells another, or a breaker controller, that something must happen right now. GOOSE data is embedded directly in Ethernet frames, published to multicast addresses, prioritised with VLAN tags and retransmitted at increasing intervals so it is not lost, with a state number that tells subscribers whether a message is new (Wikipedia, Generic Substation Events). The time budget is tiny. Trip messages are expected to arrive within about 3 to 4 milliseconds. In their standard form GOOSE messages carry no built-in security, and because they run at Ethernet layer 2 they cannot use the IP-level protection available to MMS (Herath et al., CIGRE 2025). We come back to what that means, and how IEC 62351 fixes it, in the substation section.
Meters and the edge
At the customer end sit smart meters, part of an advanced metering infrastructure (AMI). Unlike older automatic meter reading, AMI is two-way: the utility can read the meter and also send commands to it, including price signals, demand response requests and remote service disconnection. Meters reach the utility over power line communication, RF mesh or cellular links, often through data concentrators, and a head-end system manages them, including configuration and firmware updates (Wikipedia, Smart meter). The DLMS/COSEM protocol family, standardised as IEC 62056, is the most widely adopted for meter communication.
Beyond the meter there is now a great deal of distributed energy: rooftop solar, home and commercial batteries, and EV chargers. Each has an inverter or controller that decides how much power to push into or pull from the grid, a communication dongle or gateway, and usually a vendor cloud and a mobile app. That cloud can change settings across many thousands of sites at once, which is why it matters to the grid.
What the software stack looks like
Put together, a utility runs Windows and Linux servers in its control centres, embedded firmware in relays, RTUs, gateways and meters, configuration and engineering software on workstations, and a fleet of cloud and mobile apps for customers and field staff. Almost none of it was written by the utility. It came from equipment makers, who in turn built it from real time operating systems, embedded Linux, protocol stacks, cryptographic libraries and open source packages. That supply chain is where this guide ends up.
What is the attack surface of the grid?
The grid's attack surface has three bands. On the business side, email, corporate IT, vendor remote access and internet facing firewalls. On the operations side, control centre HMIs, engineering workstations, substation gateways and substation LANs. At the edge, meters, inverter clouds and apps, remote links, and the firmware and updates every device depends on.
Figure 2. Ten entry points seen in documented incidents and research.
The business side is the usual first step. Phishing, stolen passwords and unpatched internet facing devices are how attackers got into every documented case that ended in a grid outage. Vendor and contractor remote access deserves its own line, because it is designed to cross from the outside world into operational networks.
The operations side is where the damage happens. Control centre HMIs can open breakers across a region. Engineering workstations hold relay settings and the configuration files that define how a substation behaves. Substation gateways and serial-to-Ethernet converters translate control commands for older field equipment. The substation LAN carries GOOSE, MMS, DNP3 and IEC 104 traffic that, in most installations, does not authenticate its sender.
The edge keeps growing. Meters number in the millions, inverter fleets are managed through public clouds and phone apps, remote sites use cellular and satellite links, and every device depends on firmware and software updates from a maker that the utility does not control.
What does an attacker need to disrupt the grid?
Disrupting the grid takes access, knowledge and time. An attacker needs a path into operational networks, an understanding of the specific control systems and equipment in use, and usually weeks or months of quiet reconnaissance. Smaller harms, such as meter fraud or hijacked solar monitors, need far less.
The documented cases show the same needs again and again.
- A way in. Phishing emails, stolen VPN credentials, unpatched firewalls and remote access tools are the common routes.
- A route to operations. Business networks and control networks are supposed to be separate. Attackers look for jump paths: VPNs that land inside control networks, dual-homed servers, shared accounts and vendor connections.
- Knowledge of the specific environment. The three Ukrainian distribution companies hit in 2015 used three different distribution management systems, and the attackers had to learn each one (E-ISAC and SANS ICS).
- Equipment to test against. The same report assessed with high confidence that the attackers tested their malicious firmware before using it, which implies access to matching hardware.
- Time. In 2015 the activity ran from at least March to December.
Attacks at the edge need much less. Meter fraud has been done with an optical probe and a laptop. Hijacking internet exposed solar monitoring devices has been done with a known command injection flaw. Account takeover in an inverter cloud needs only a browser and a flawed API. The impact of each single case is small. The worry is scale.
What happened in the 2015 Ukrainian grid outage?
On 23 December 2015, three Ukrainian distribution companies lost control of their systems. Attackers who had entered through phishing months earlier used the operators' own SCADA screens to open breakers, cutting power to about 225,000 customers. They also overwrote substation gateway firmware and wiped computers. Crews restored power manually within hours.
This was the first publicly acknowledged cyber attack to cause a power outage, and it remains the best documented. The account below follows the joint analysis by the Electricity Information Sharing and Analysis Center (E-ISAC) and the SANS Industrial Control Systems team, published in March 2016, which explicitly does not focus on attribution (E-ISAC and SANS ICS). We follow the same approach: what happened technically, and what it teaches.
Figure 3. The 2015 outage step by step, with the controls that would have changed the outcome.
Step 1: phishing into the business network
The attackers sent targeted emails to specific staff with Microsoft Office attachments that looked as if they came from a trusted source. Opening the document showed a prompt encouraging the reader to enable macros. Enabling them installed the BlackEnergy 3 malware on the business network, giving the attackers a foothold.
Why it worked: the documents looked legitimate and macros were allowed. What would have helped: blocking macros from the internet, sandboxing attachments, and assuming the business network can be compromised at any time.
Step 2: credential theft and the road to operations
From that foothold, the attackers harvested credentials, including with keystroke loggers, and studied the network. They found VPN connections that led from the business network into the industrial control network, and they used them. The report lists the use of VPNs to enter the ICS network as one of the attack's core components.
Why it worked: the VPN accepted a password alone. The report's first mitigation for this stage is two-factor authentication with user tokens, followed by trusted jump hosts with network access control and VPN configurations that disable split tunnelling.
Step 3: learning three different control systems
The three companies used three different distribution management systems. No vendor exploit was used against any of them; the report says the products "were simply abused with direct access". The attackers learned how each system's native screens and controls worked, and discovered the field devices, including the serial-to-Ethernet converters that pass commands from the SCADA network to substation equipment.
They also developed malicious firmware for those converters. The firmware was consistent across devices and uploaded to several sites within short periods, which the analysts read as preparation and testing done well before the attack.
Step 4: opening the breakers
On the afternoon of 23 December, the attackers took control of operator workstations, in some cases using remote administration tools already installed on them, and locked operators out. They then used the HMIs in the SCADA environment to open breakers. At least 27 substations were taken offline across the three companies, and the analysts assessed that the true number was probably higher. One company's own update said seven 110 kV and twenty-three 35 kV substations were disconnected for three hours. The coordinated actions at the three companies happened within about 30 minutes of each other.
Why it worked: opening breakers from the control centre is a feature, not a flaw. Dragos and SANS put it plainly ten years later: opening circuit breakers through the DMS "isn't a vulnerability, it's a feature", and the vulnerability was that an adversary had access and operators had no way to detect the takeover (Dragos).
Step 5: blocking recovery
At the same time, the attackers worked to slow restoration:
- They uploaded the malicious firmware to the serial-to-Ethernet gateways, so that even if operator workstations were recovered, remote commands could not reach the substations. The analysts called this "blowing the bridges", and some devices were left unrecoverable.
- A modified version of the KillDisk wiper erased the master boot record on workstations and servers, and on an HMI card inside an RTU.
- In at least one company, a network-connected uninterruptible power supply (UPS) had been reconfigured so that the outage was followed by a loss of power to the company's own buildings or data rooms.
- A flood of thousands of calls hit a call centre, so customers could not report outages.
Step 6: manual restoration
The companies switched to manual operations, sending crews to substations to close breakers by hand. Power came back after several hours. Dragos notes that the roughly six hour restoration is often misread as a defensive success. It happened because Ukrainian operators still knew how to run the grid manually, and the companies lost automation capabilities for almost a year afterwards (Dragos).
Impact and lessons
The analysts rated the event as low on a national power system scale, since it affected a small share of consumers for a limited time, while noting that the affected companies would rightly rate it as critical. The lessons are what made it famous:
- Segmentation alone was not enough. The companies had firewalls, antivirus, patching and access controls. Attackers used legitimate paths that crossed those boundaries.
- The control system did what it was told. Area of responsibility limits, which restrict which breakers each operator station can command, would have narrowed the blast radius. So would confirmation steps in the HMI.
- Field device firmware is a target. Remote firmware updates that were left enabled let the attackers brick gateways. The report recommends disabling remote management of field devices when it is not needed, and keeping a vaulted copy of known good firmware, project files and logic.
- Building systems belong outside the control network. The UPS should never have been reachable from it.
- Manual operation is a safety system. Practise it.
What happened in 2016, and how did Industroyer work?
In December 2016 a transmission substation near Kyiv lost power for about an hour. Investigators later found Industroyer, malware with modules that spoke four grid protocols directly, so it could command switches and breakers itself. It also carried a wiper and a tool aimed at protection relays. Its lesson is that grid protocols obey whoever can reach them.
The 2016 event
On the night of 17 to 18 December 2016, an outage at the Pivnichna transmission substation caused blackouts in parts of Kyiv and the surrounding region. Power was fully restored after just over an hour. The national transmission company, Ukrenergo, said its preliminary analysis showed that workstations and SCADA servers had been disrupted by malware that let attackers control internal systems remotely (SecurityWeek).
Industroyer, technically
In June 2017 ESET published its analysis of the malware it named Industroyer, whose activation timestamp matched the date of the outage (ESET). Dragos analysed the same toolkit under the name CRASHOVERRIDE. Its design is what made it notable.
Figure 4. Industroyer's modular design, and how 2016 differed from 2015.
- A main backdoor controlled the other components and communicated with its operators, with a secondary backdoor disguised as an ordinary Windows program as a fallback.
- Four payload modules each spoke one protocol: IEC 60870-5-101, IEC 60870-5-104, IEC 61850 and OLE for Process Control Data Access (OPC DA). Each worked in stages, first mapping the network and then issuing commands suited to the devices it found.
- A data wiper erased registry keys and overwrote files so systems could not boot, which slows recovery.
- A custom port scanner found relevant machines.
- A denial of service tool targeted CVE-2015-5374, a known flaw in the EN100 Ethernet module of Siemens SIPROTEC protection relays that could make them unresponsive. The CVE record, published in July 2015, lists the fixed firmware versions.
ESET's central point is worth quoting. Industroyer's danger "lies in the fact that it uses protocols in the way they were designed to be used." The protocols were not designed with security in mind, so the malware did not need a protocol vulnerability. It needed network access and a correct implementation of the protocol.
Protection relays were part of the design. Industroyer's denial of service tool targeted relays, and Dragos notes that the toolkit could load different exploits for different protection relays, and that adversaries have since targeted protection relays and safety systems not only to cause outages but to create conditions for equipment damage. A relay that has stopped responding cannot isolate a fault when operators restore power. Ten years on, Dragos summarised the shift this way: what took a team of people and 45 minutes of manual operations in 2015 became something that could be executed in 45 seconds in 2016, with protocol modules that could be swapped for different targets (Dragos).
A narrower variant in 2022
In April 2022, ESET and Ukrainian incident responders found a new variant at a Ukrainian energy company, which they named Industroyer2. It implemented only IEC 60870-5-104, with its target settings hardcoded into the binary, and it was scheduled to run against high voltage substations on 8 April 2022 alongside wipers for Windows, Linux and Solaris systems. The responders reported that the plan was prevented (ESET). The technical lesson is the same: a small, well understood protocol client, pointed at devices that accept unauthenticated commands, is enough.
How to detect and defend against protocol aware malware
- Know your normal. Substation traffic is highly regular. A new host sending IEC 104 or MMS control commands, commands at unusual times, or a burst of breaker operations across many sites are strong signals. Passive network monitoring that understands these protocols is the main way to see them.
- Restrict who can send control commands. Allow lists at the substation gateway and on the station network should permit control traffic only from known control centre hosts.
- Turn on protocol authentication where the equipment supports it. DNP3 Secure Authentication, IEC 62351 profiles for MMS and IEC 104, and GOOSE authentication all raise the bar.
- Patch relays, and keep the patch evidence. CVE-2015-5374 had a fix more than a year before Industroyer used it.
- Keep protection independent. Relays that act locally, with settings under change control, are the last line of defence when the control centre cannot be trusted.
Can a cyber attack physically damage grid equipment?
In principle yes, which is why protection systems exist. The 2007 Aurora generator test showed that rapidly opening and closing a breaker out of phase with the grid can wreck a generator. The defences are well known: synchronism-check functions in relays, reclosing supervision, and protection settings that cannot be changed remotely without review.
The best known demonstration is the 2007 Aurora generator test. Researchers connected a 2.25 MW diesel generator to a substation and used a computer to open and close its breakers out of phase with the rest of the grid. Each out-of-phase reconnection subjected the machine to abnormal torque, and the generator was destroyed within about three minutes (Wikipedia, Aurora Generator Test). We keep this high level on purpose. The concept is what matters for defenders: the grid's own switching equipment can stress rotating machines if it is commanded in the wrong sequence, faster than protection reacts.
The mitigations are mature. A synchronism-check function, the sync-check relay known by the device number 25, refuses to close a breaker unless voltage, frequency and phase angle on both sides are within a set window. Time delays on reclosing, and protection that blocks reclosing after abnormal events, add margin. And the settings that control all of this live in the relays themselves, which is why relay settings and firmware deserve the same change control as anything else in a substation.
Protection relays are the reason grid cyber attacks so far have caused outages rather than physical destruction. Their job is to isolate faults automatically and quickly, whatever the control centre is doing. An attacker who wants damage has to defeat protection, which is why Industroyer carried a tool aimed at relays. For defenders, that makes relays and their engineering workstations the most sensitive assets in the substation.
How are substations and IEC 61850 networks attacked?
Substations are attacked from the inside. Once on the station network, an attacker can send MMS commands to IEDs, inject or replay GOOSE messages, tamper with relay settings from an engineering workstation, or exploit flaws in gateways and protocol stacks. IEC 62351 adds authentication, and monitoring catches what it misses.
Figure 5. IEC 61850 inside a digital substation, and where IEC 62351 adds security.
GOOSE spoofing and replay
GOOSE was designed for speed on a trusted network. A subscriber accepts a message if it arrives on the right multicast address with the right identifiers and a plausible state and sequence number. In its standard form there is no cryptographic proof of who sent it. Researchers have described three main attack classes: replay, resending a captured message; masquerade, forging a message that claims to come from a legitimate publisher; and flooding, overwhelming subscribers (Herath et al., CIGRE 2025).
What the attacker needs: a foothold on the substation LAN itself, plus knowledge of the substation configuration, which is described in Substation Configuration Language (SCL) files kept on engineering workstations. Impact: a false trip or a blocked trip in one substation. Why it is not the headline risk: it requires deep, local access, and the impact is contained to that site.
Defence. IEC 62351-6 defines security for IEC 61850 profiles. Its 2020 edition recommends message authentication codes (MACs) with pre-shared keys for GOOSE and Sampled Values, appended as an extension to each message, and adds checks on state and sequence numbers to catch replays. In a 2025 evaluation, researchers found MAC verification plus intrusion detection added well under the 3 millisecond budget required for the fastest GOOSE messages (Herath et al.). Where devices do not yet support it, a GOOSE-aware intrusion detection system that tracks state and sequence numbers and timing is the practical fallback.
MMS commands and engineering workstations
MMS is the client and server side of IEC 61850: reading values, changing settings and issuing controls. An attacker with access to the station bus and a correct MMS client can do what the local HMI can do. Industroyer's IEC 61850 module was exactly that.
Engineering workstations are a softer target. They run the vendor software that sets relay protection logic and pushes firmware. In 2023, Nozomi Networks Labs disclosed nine vulnerabilities in Schweitzer Engineering Laboratories engineering software, the most severe of which could allow remote code execution on the workstation, including through a crafted configuration file. Earlier that year it disclosed 19 vulnerabilities in the web interface of SEL's Real-Time Automation Controller (RTAC) platform, CVE-2023-31148 through CVE-2023-31166, which covered issues from improper input validation to passwords stored in a recoverable format (Nozomi Networks Labs). SEL shipped fixes. The point is not one vendor. It is that the tools used to configure protection are themselves software with bugs.
Insecure by design
In 2022, Forescout's OT:ICEFALL research disclosed 56 vulnerabilities across devices from ten operational technology vendors and grouped them into insecure engineering protocols, weak cryptography or broken authentication, insecure firmware updates, and remote code execution through native functionality (Forescout). Many such issues had historically gone without CVE numbers because they were considered design choices rather than bugs. Dragos reports a related statistic from its own analysis: 64% of ICS vulnerability patches do not actually eliminate risk, because the component being patched was insecure by design, and 72% of ICS vulnerabilities came with no mitigation guidance beyond patching (Dragos).
Protocol stacks are software too
Many products implement IEC 61850 using shared libraries. One widely used option is libiec61850, an open source (GPLv3) C library from MZ Automation that implements MMS, GOOSE, SV and R-Session, with commercial licences for product makers (libiec61850 on GitHub). Like any parser that reads network input, it has had memory safety bugs. CVE-2022-2972 is a stack-based buffer overflow in versions 1.4 and earlier that could allow a crash or remote code execution, and CVE-2022-21159, found by Cisco Talos, is a denial of service in presentation layer parsing in version 1.5.0. Both were fixed. But a fix in the library only protects a substation when every product that embeds it ships an update, and that is a supply chain question.
How does vendor remote access put the grid at risk?
Vendor remote access puts the grid at risk because it is a deliberate path from the internet into operational networks. Attackers use stolen vendor credentials, exposed VPN gateways and unpatched firewalls, and sometimes reach devices the utility did not know it had. Jump hosts, multi-factor login, time limits and monitoring close most of the gap.
Equipment makers and integrators support what they sell. That often means a VPN account, a remote support tool on a workstation, or a cellular router in a cabinet. Every one of those is a door.
Figure 6. Purdue levels as IEC 62443 zones and conduits. Vendors enter through the DMZ only.
The Ukrainian VPNs
In 2015 the attackers used VPN access into the control network with harvested credentials. The E-ISAC and SANS analysts warned that the next attack might look for point-to-point VPNs at trusted third party networks, or remote support connections with split tunnelling enabled. Their recommendations are a short manual for vendor access: ask why each trusted path exists and remove those without a clear need; force remote VPNs through a dedicated remote access DMZ; use time-of-use access that disconnects automatically after a set period; and monitor connection duration, count and timing for anomalies (E-ISAC and SANS ICS).
Danish energy companies, May 2023
In May 2023, 22 Danish energy sector companies were breached within a few days, almost all through Zyxel firewalls with an unpatched command injection vulnerability, CVE-2023-28771, which let an unauthenticated attacker run operating system commands. The report by SektorCERT, the Danish critical infrastructure security organisation, explained why the patch had not been applied, in words most utilities will recognise. Some members assumed a new firewall must be up to date. Some thought the vendor handled updates. Some had opted out because the supplier charged to install them. And some did not know they had the device at all, including one organisation whose Zyxel firewall had been installed by a third party supplier when it set up a camera system (The Register).
Several companies went into island mode, disconnecting from the internet and cutting non-essential connections while they cleaned up. In one case, a compromised firewall was drafted into a botnet used for denial of service attacks elsewhere and had to be replaced entirely. In another, the company lost visibility of three remote sites and sent staff to drive out and operate them manually. The report found no significant impact on the operation of Danish critical infrastructure.
sPower, March 2019
In March 2019, attackers exploited a known vulnerability in an unpatched Cisco firewall at sPower, a US wind and solar operator, causing the firewall to reboot repeatedly over about 12 hours. The attack disrupted the company's visibility into about 500 MW of wind and solar generation in California, Utah and Wyoming. There was no loss of generation, and reporting suggested the attackers may not even have known the firewall was connected to energy assets (Utility Dive). The fix was a firmware update to the firewall.
Wind turbines and a satellite link, February 2022
Remote links can fail without anyone targeting the grid. On 24 February 2022, a cyber attack on the KA-SAT satellite network disrupted around 30,000 satellite terminals across Europe. Among them were the remote monitoring links for 5,800 Enercon wind turbines in central Europe with more than 10 gigawatts of capacity. Enercon said there was never a risk to the turbines and that grid operators kept unrestricted access to control their output, but SCADA monitoring of the affected turbines was lost. Enercon replaced damaged modems, moved some sites to LTE, and by 19 April had more than 95% back online. It then began offering LTE backup retrofit kits, noting that many wind farms had no backup communication link (REVE).
How to detect and defend
- One door, watched. All vendor and remote staff access lands on a jump host in an operations DMZ, never directly on a control network.
- Multi-factor login and named accounts. No shared vendor passwords. NERC CIP-005 requires multi-factor authentication for interactive remote access to high and medium impact systems, through an intermediate system (NERC CIP-005-7).
- Time limited, approved sessions. Access is enabled for a ticket and switched off afterwards. Sessions are recorded.
- Find the doors you do not know about. Regular asset discovery, including cellular routers and firewalls installed by contractors.
- Patch edge devices first. Firewalls and VPN gateways are internet facing by definition, and known exploited vulnerabilities in them are fixed fast or the device is isolated.
- Plan for lost links. Backup communication paths and a practised drive-out procedure turn a lost link into an inconvenience.
How do smart meters get hacked?
Smart meters have been hacked mostly for fraud, by reprogramming them through optical maintenance ports to under-record usage. Researchers have also shown flaws that could let an attacker control meters over the utility network. The bigger concern is the remote disconnect switch: a head-end compromise could cut many homes at once.
Fraud through the optical port
The best documented real world case is fraud. KrebsOnSecurity reported in 2012 on a law enforcement bulletin describing how a utility in Puerto Rico had asked for help in 2009 with widespread power theft linked to its smart meter rollout. Insiders and former employees of the meter maker were reportedly charging 300 to 1,000 US dollars to reprogram residential meters. They connected a laptop to the meter's optical port, the infrared interface intended for field technicians, and changed how consumption was recorded, typically cutting a bill by 50 to 75%. A spot check found about 10% of meters altered, and losses were estimated at up to 400 million US dollars a year. The method required physical access and left the meter intact (KrebsOnSecurity).
Why it worked: the credentials needed to change meter settings were poorly protected, and anyone with the right probe and software could use the maintenance interface. Defence: per-meter credentials, authenticated configuration changes, tamper logging, and analytics that compare a meter's readings with neighbouring meters and with transformer-level measurements.
Control over the meter network
At Black Hat Europe in 2014, researchers Javier Vazquez Vidal and Alberto Garcia Illera described reverse engineering smart meters deployed in Spain. Working on their own small network of meters, they reported that poorly protected credentials and keys inside the devices could let an attacker take control of a meter over the power line network, impersonate other meters, tamper with readings or cut supply. They withheld technical detail until the problem was fixed, and the utility said it was improving the devices' security (The Hacker News).
The off switch
In 2010, Ross Anderson and Shailendra Fuloria of Cambridge argued that the remote disconnect switch in smart meters creates a security problem of a new kind and scale. If an attacker could command millions of meters to switch off, and perhaps change their keys so they could not easily be switched back on, the result would be a mass outage that no physical attack could easily match. They argued for safeguards before large rollouts, such as a delay before a disconnection takes effect and requiring more than one party to cooperate before anyone can take control of a meter (Anderson and Fuloria).
No such event has been reported. The argument still shapes how AMI should be built: rate limits and approval workflows on mass disconnect commands at the head-end, strong authentication between head-end and meters, unique keys per meter, signed firmware, and a head-end treated as a high value control system rather than a billing server.
How do solar inverters, batteries and EV chargers get hacked?
Most published solar flaws are in the clouds, apps and communication dongles that manage inverters, not the power electronics. Researchers have found APIs that expose other users' systems, hard coded credentials and remote code execution. One flaw can reach thousands of sites, and researchers warn that coordinated control of a large fleet could stress grid stability.
Distributed energy resources (DER) are now a large share of generation in many places. An inverter converts DC from panels or batteries into grid-synchronised AC, and its settings decide how it behaves when frequency or voltage moves. Those settings can usually be changed remotely, by the owner through an app, by the installer through a portal, or by the maker through its cloud.
Figure 7. How inverter fleets are reached through clouds, apps and dongles.
SUN:DOWN, 2025
In March 2025, Forescout's Vedere Labs published SUN:DOWN, a study of solar power system security that disclosed 46 new vulnerabilities in products from three of the top ten inverter vendors: Sungrow, Growatt and SMA (Forescout). The vulnerabilities, as reported, included:
- Growatt cloud APIs that let an unauthenticated attacker obtain lists of other users' plants and devices, and smart meter serial numbers, leading to device and account takeover, plus an EV charging endpoint that exposed charger information and allowed remote configuration.
- Sungrow's Android app using an insecure encryption key for client data and ignoring certificate errors, which opens the door to interception, along with flaws in how Sungrow devices handle MQTT messages that could lead to remote code execution or denial of service. Forescout described a path from serial numbers harvested through insecure direct object references, to hard coded credentials on communication dongles, to full inverter takeover.
- A flaw in SMA's Sunny Portal web service that allowed uploaded files to be executed by the server. SMA said the issue affected Sunny Portal, not its inverters, and was resolved on 19 December 2024 with no evidence of exploitation (The Hacker News).
Forescout also reviewed earlier disclosures and found that 80% of 93 previously disclosed solar vulnerabilities were rated high or critical. All three vendors fixed the newly reported issues after coordinated disclosure. Forescout's grid scenario is the important part: an attacker controlling a large fleet could coordinate changes in output to destabilise the grid, potentially triggering emergency measures. Its recommendations for owners and operators were strict procurement requirements, regular risk assessments, full network visibility into these devices, and segmentation into monitored sub-networks.
Solarman and Deye, 2024
In 2024, Bitdefender reported vulnerabilities in the Solarman and Deye platforms, which Bitdefender said coordinate millions of installations with about 195 GW of output, around 20% of global solar production by its estimate. The issues included an authorisation token endpoint that could be made to issue tokens for any account, and tokens from one platform being accepted by the other. Bitdefender said an attacker could have controlled inverter settings, and the vendors shipped fixes by mid-July 2024 after disclosure in May (Bitdefender).
Enphase IQ Gateway, 2024
Researchers at the Dutch Institute for Vulnerability Disclosure (DIVD) found six vulnerabilities in Enphase IQ Gateway devices, formerly called Envoy, in versions 4.x to 8.x. Three could be combined into unauthenticated remote command execution. Enphase reported fixes deployed in July 2024, from version 8.2.4225, and pushes firmware to devices itself. DIVD's advice to owners was simple: do not expose the gateway to untrusted networks, and keep it behind NAT if it needs internet access (DIVD).
SolarView monitoring devices, 2024
In 2024, Japanese media reported that 800 SolarView Compact remote monitoring devices made by Contec, installed at solar generation sites, had been hijacked and used in bank account theft. The attackers were presumed to have exploited CVE-2022-29303, a command injection flaw in the device's web interface that had been patched in 2023 and was already being exploited to spread a botnet. The incident did not threaten power system operations, but researchers interviewed by CSO noted that a monitoring device on a plant network gives an attacker local network access, and some such devices can shut a solar installation down (CSO Online).
The demand side: high wattage devices
The same fleet logic applies to load. In 2018, Princeton researchers Saleh Soltan, Prateek Mittal and H. Vincent Poor showed in simulation that a botnet of high wattage devices such as air conditioners and heaters, switched in unison, could manipulate demand enough to disrupt a power grid. They called these Manipulation of demand via IoT (MadIoT) attacks (USENIX Security 2018). Smart thermostats, home batteries and EV chargers all fall into this class, which is why their cloud security matters beyond the individual home. Our guide on how cars and EV chargers get hacked covers charger firmware and OCPP backends in detail.
Detection and defence at the edge
- For owners: keep gateways and inverters off the open internet, change default passwords, leave updates on, use a strong unique app password with multi-factor login, and remove installer access you no longer need.
- For utilities and aggregators: treat DER clouds as part of the grid. Require security commitments in interconnection and procurement agreements, monitor aggregate behaviour for coordinated setting changes, and design grid codes so that inverters ride through disturbances rather than all tripping at once.
- For makers: authorise every API call against the object it touches, enforce TLS certificate validation in apps, remove hard coded credentials, rate-limit and approve fleet-wide setting changes, and sign firmware.
How should you weigh these risks?
Weigh grid risks by impact on supply against the effort an attacker needs. Control centre takeovers and protocol aware malware are rare and need months of preparation, but they cause outages. Edge attacks such as meter fraud and hijacked solar monitors are common and cheap but local. Fleet-wide cloud flaws in inverters sit in between and are growing fastest.
Figure 8. The events and research that shaped grid security, 2007 to 2025.
Two observations come out of the timeline. First, the attacks that actually cut power were long, prepared operations against utilities, and their decisive step used legitimate control functions. Second, the volume of published vulnerabilities has shifted toward the edge of the grid, where devices are more numerous, more connected to the internet and managed through consumer-grade clouds.
Figure 9. Impact against attacker effort, based on the cases in this guide. Placement is a qualitative judgement.
A practical reading for a utility:
- Top right (rare, prepared, severe): control centre takeover through HMIs, protocol aware malware and relay attacks. Defend with segmentation, remote access control, monitoring that understands grid protocols, independent protection and practised manual operation.
- Top left (cheap and disruptive): account and API takeover of inverter fleets. You may not own these systems, but your grid absorbs their behaviour. Defend through procurement, interconnection requirements and aggregate monitoring.
- Bottom left (common, local, mostly financial): meter fraud, hijacked monitors and edge firewall denial of service. Defend with patching, credential hygiene and analytics.
- Bottom right (hard, contained): GOOSE spoofing inside one substation. Defend with IEC 62351-6 where supported and substation intrusion detection.
How do you detect attacks on grid systems?
Detect grid attacks by watching the operational network itself. Passive monitoring that understands DNP3, IEC 104, MMS and GOOSE can flag new hosts, unusual commands and firmware writes. Combine it with SCADA application logs, remote access records, asset inventory and operators who report anything strange.
The 2015 attackers spent months inside the network. Every stage left signals that someone could have seen. The challenge, as Dragos puts it, is that much OT forensic data is transient: if network traffic between control systems is not collected ahead of time, it is gone after the incident (Dragos).
What to watch
- Remote access. VPN and jump host logins outside maintenance windows, from new locations, or lasting unusually long. The E-ISAC and SANS analysts recommended monitoring connection duration, count and timing at a dedicated remote access DMZ.
- New talkers on control networks. A workstation that has never sent IEC 104, DNP3 or MMS commands and starts doing so. A new device appearing on a substation LAN.
- Command patterns. Breaker operations outside switching orders, simultaneous operations across many substations, or control commands at times when nobody is dispatching.
- Firmware and settings changes. Any firmware upload to a relay, RTU or gateway, and any relay setting change, should match an approved change ticket. CIP-010 requires a baseline that includes firmware versions, installed software and open ports, and monitoring for deviations from it (NERC CIP-010-4).
- GOOSE anomalies. State and sequence numbers that jump or repeat, publishers on the wrong port, or message rates that change.
- Edge device health. Firewall reboots, configuration exports and outbound connections from firewalls and VPN gateways, which were the signals in the sPower and Danish cases.
- The operators. The 2015 analysts advised defenders to "talk to operators and ask about abnormal occurrences". A cursor that moves by itself is a security event.
Who does it
OT network monitoring is a specialised field with dedicated vendors such as Dragos, Nozomi Networks and Forescout, along with open source tools for protocol parsing. In North America, NERC's newer standard CIP-015-1 makes internal network security monitoring a requirement for applicable systems, with the stated purpose of improving the probability of detecting anomalous or unauthorised network activity (NERC CIP-015-1). Dragos recommends that organisations outside NERC's scope treat it as a model.
Respond in a way that keeps the lights on
Detection is only useful if operators can act on it. The E-ISAC and SANS report describes moving to a conservative operations mode in which control from untrusted assets is paused, remote connections are cut, and incident response runs before control is restored to known good assets. Dragos adds that utilities may choose to de-energise a line or de-rate a unit if they detect manipulation, in the same way they plan outages in wildfire conditions. Both require a plan written and practised before the day it is needed.
What do IEC 62443, IEC 62351 and NERC CIP require?
IEC 62443 is the general standard family for industrial automation security, covering asset owners, system integrators and product makers, with zones, conduits and security levels. IEC 62351 secures the power system protocols themselves. NERC CIP is the mandatory reliability standard set for the North American bulk electric system.
IEC 62443
IEC 62443 is a series of standards for the security of industrial automation and control systems. It is organised into general concepts, policies and procedures for asset owners, system requirements, and component requirements (Wikipedia, IEC 62443). The parts most relevant here:
- Zones and conduits. Assets with common security requirements are grouped into zones. Conduits are the communication paths between zones, with security functions that let zones of different levels coexist. IEC 62443-3-2 describes how to partition a system into zones and conduits and set a target security level for each from a risk assessment, for example by separating business and control assets, safety related assets, temporarily connected devices, wireless devices and devices reached through external networks.
- Security levels. Four levels, from protection against accidental misuse (SL 1) to protection against sophisticated attackers with extensive resources and automation specific knowledge (SL 4), are used to state requirements for systems (IEC 62443-3-3) and components (IEC 62443-4-2).
- Secure development. IEC 62443-4-1 sets requirements for how product makers develop, test and patch products. Certification schemes such as ISASecure assess components and development processes against 4-1 and 4-2.
For utilities, IEC 62443 provides the vocabulary for segmentation (Figure 6). For equipment makers, 4-1 and 4-2 are what customers increasingly ask for in procurement.
IEC 62351
IEC 62351 is the security series for the power system communication protocols of IEC Technical Committee 57, including IEC 60870-5, IEC 60870-6, IEC 61850, IEC 61970 and IEC 61968 (Wikipedia, IEC 62351). Its goals include authentication of data transfer, prevention of eavesdropping, playback and spoofing, and intrusion detection. Key parts:
- 62351-3: TLS and X.509 certificate based node authentication for TCP/IP profiles.
- 62351-4: security for MMS based profiles, including IEC 61850 client and server traffic.
- 62351-5: security for IEC 60870-5 and its derivatives. DNP3 Secure Authentication is compliant with it.
- 62351-6: security for IEC 61850 profiles, including GOOSE and Sampled Values authentication.
- 62351-8: role based access control for users and automated agents.
- 62351-9: key and certificate management across their life cycle.
The standard exists. The gap is deployment: many installed devices do not support it, and turning it on requires key management that utilities have to build.
NERC CIP
The North American Electric Reliability Corporation (NERC) Critical Infrastructure Protection standards are mandatory for owners and operators of the bulk electric system in North America. They are also, as Dragos notes, a comprehensive model that utilities elsewhere can learn from. The core set:
- CIP-002: identify and categorise BES Cyber Systems as high, medium or low impact (CIP-002-5.1a).
- CIP-003: security management controls, including for low impact systems (CIP-003-9).
- CIP-005: the Electronic Security Perimeter, with interactive remote access through an intermediate system, encryption to that system, multi-factor authentication, and the ability to detect and disable active vendor remote access (CIP-005-7).
- CIP-007: system security management, including a patch management process that evaluates newly released security patches at least once every 35 calendar days (CIP-007-6).
- CIP-010: configuration change management, with baselines covering operating system or firmware version, commercial and open source software, custom software, open ports and applied patches (CIP-010-4).
- CIP-013: supply chain risk management. Procurement processes must address vendor incident notification, coordination of incident response, notice when vendor access should be revoked, disclosure of known vulnerabilities, verification of software integrity and authenticity for all software and patches, and coordination of vendor remote access (CIP-013-2).
- CIP-015: internal network security monitoring (CIP-015-1).
Read CIP-010 and CIP-013 together and the supply chain shape is obvious. A utility must know exactly what software and firmware is on each system, must hear from vendors about vulnerabilities in it, and must verify that every update is genuine. Those are SBOM, vulnerability management and signing problems.
Why is grid security a software supply chain problem?
Grid security is a supply chain problem because utilities run software they did not write. Relays, RTUs, gateways, meters, inverters and control centre systems are built by equipment makers from operating systems, protocol stacks and open source libraries. A vulnerability anywhere in that chain becomes the utility's risk, and only the maker can ship the fix.
Look back over this guide. The gateway firmware overwritten in 2015 came from a device maker. The relay flaw Industroyer targeted was in a vendor's Ethernet module. The firewalls in Denmark and at sPower were bought from vendors and, in one case, installed by a contractor the utility did not know had done so. The SEL engineering tools, the solar clouds, the Enphase gateway and the SolarView monitor were all vendor software. The libiec61850 bugs sat in an open source library that many products embed.
The chain
- Chip and platform suppliers provide processors, real time operating systems, embedded Linux distributions, board support packages and SDKs.
- Protocol and library suppliers provide IEC 61850, DNP3 and IEC 104 stacks, TLS libraries, web servers and MQTT clients, many of them open source.
- Equipment makers build relays, RTUs, meters, inverters and gateways on top, add their own code, and ship firmware.
- Software makers ship SCADA, EMS, DMS, head-end and engineering software, plus mobile apps and clouds.
- Integrators assemble and configure systems for each utility, often with their own remote access.
- Utilities and operators run all of it for decades.
Grid equipment stays in service for many years, far longer than a phone or a laptop. Over that time, components inside it will have vulnerabilities disclosed. The utility can only act on them if it knows what is inside, and only the maker can rebuild the firmware.
Why it is hard
- Opacity. Firmware is delivered as a binary. Without an SBOM, a utility cannot tell whether a device contains an affected library version.
- Insecure by design. As OT:ICEFALL and Dragos's statistics show, some risk is in the design rather than a bug, so there is no patch to wait for.
- Slow updates. Updating a relay can mean a planned outage, testing and sometimes requalification, so field devices are often patched far less often than IT systems.
- Long tails. A fix in an upstream open source library only helps once every maker that embeds it ships a new release, and every utility installs it.
What good looks like
- SBOMs for every firmware release, application and image, delivered with the product and kept current.
- Vulnerability disclosure from vendors, as CIP-013 requires, with clear statements of which products and versions are affected and whether the vulnerable code is reachable.
- Signed firmware and verified updates, so that the 2015 trick of uploading malicious firmware to a gateway fails at the device.
- A way to fix open source components fast, even when the upstream project has not released a fix or the fix sits in a version that would force a larger upgrade.
Our deeper guides on software supply chain security for electric utilities, NERC CIP and the utility software supply chain, assessing software suppliers with C2M2 at a US electric utility and eliminating vulnerability exposure in a utility OT vendor stack go further.
How should utilities and grid operators secure their systems?
Utilities should know every asset and its firmware, segment networks using Purdue levels and IEC 62443 zones, force all remote access through monitored jump hosts with multi-factor login, monitor OT protocols, keep offline backups of firmware and settings, practise manual operation, and hold vendors to supply chain requirements.
Figure 10. Seven defence layers. Each one limits what an attacker who beats the last can do.
1. Asset inventory
- Keep a live inventory of every relay, RTU, gateway, serial converter, firewall, switch, server, workstation, meter type and inverter fleet, with make, model, firmware version and location.
- Include devices installed by contractors and integrators. The Danish case shows why.
- Link each asset to its SBOM where the vendor provides one, and ask for one where it does not.
2. Segmentation with the Purdue model and IEC 62443
The Purdue model describes industrial networks as levels: level 0 is the physical process, level 1 the devices that sense and act on it, level 2 the control systems and HMIs, level 3 operations systems such as historians, and level 4 business systems (Wikipedia, Purdue Enterprise Reference Architecture). Utilities usually add a demilitarised zone between levels 3 and 4.
- Treat each level, or each substation, as an IEC 62443 zone with a target security level.
- Allow traffic only through defined conduits, with allow lists by host, port and protocol.
- Allow no direct path from the business network to level 2 or below.
- Keep building systems, such as UPS units, cameras and HVAC, off control networks.
- Restrict control commands at substation gateways to known control centre hosts.
3. Secure remote access
- Route all remote access, staff and vendor, through jump hosts in an operations DMZ.
- Require multi-factor login and named accounts, with no shared vendor passwords.
- Enable vendor access per ticket, for a limited time, and record sessions.
- Disable split tunnelling on VPNs.
- Disable remote firmware updates and remote management of field devices when they are not in use.
4. Monitoring
- Deploy passive monitoring on control centre and substation networks that understands grid protocols.
- Collect SCADA application logs, not only operating system logs, and correlate them centrally.
- Alert on firmware writes, setting changes, new hosts and unusual command patterns.
- Apply area of responsibility limits in SCADA so each operator station controls only its own part of the network.
5. Protection and safety
- Keep protection relays and interlocks able to act independently of SCADA.
- Use synchronism-check on breakers that can connect two systems, and supervise reclosing.
- Put relay settings and firmware under change control, with engineering workstations treated as high value assets.
6. Backups, manual operation and incident response
- Keep offline, verified copies of known good firmware, relay settings, SCL files, PLC logic and system images.
- Practise manual operation of substations, including restoration without SCADA.
- Write and drill an incident response plan that includes a conservative operations mode, communication with neighbouring utilities, and how to restore control only to known good assets.
- Prepare for loss of communication links with backup paths and drive-out procedures.
7. Patching and the supply chain
- Patch internet facing devices, especially firewalls and VPN gateways, first and fast.
- Evaluate security patches for control systems on a fixed cycle. CIP-007 sets 35 days for evaluation.
- Require vendors to provide SBOMs, vulnerability disclosure, reachability context where they have it, signed firmware and a disclosure contact, in line with CIP-013.
- Score and monitor suppliers continuously, not only at procurement.
Requirements for distributed energy
- Write cybersecurity requirements into interconnection agreements and procurement for inverters, batteries and aggregators.
- Monitor aggregate DER behaviour for coordinated setting changes.
- Prefer grid codes and settings that make inverters ride through disturbances, so one event does not trip a whole fleet.
How should solar, battery and EV charger owners protect themselves?
Solar, battery and EV charger owners should keep inverters and gateways off the open internet, change default passwords, leave firmware updates on, use a strong unique app password with multi-factor login, put energy devices on their own network, and remove installer access they no longer need.
Figure 11. First steps for each group. The full lists are in the article.
- Keep devices off the open internet. Do not forward router ports to an inverter, gateway or monitoring device. DIVD's advice for Enphase gateways applies broadly: if a device needs internet access, keep it behind NAT (DIVD).
- Change default passwords on dongles, gateways and local web interfaces.
- Leave automatic updates on. Most fixes in this guide were delivered by the vendor's cloud or by an update the owner had to accept.
- Protect the app account. Use a unique password and turn on multi-factor login if offered. Most solar flaws in this guide were account and API problems.
- Use a separate network. A guest network or VLAN for energy devices keeps a compromised dongle away from your laptops and phones. Our guide on how routers get hacked explains how.
- Review installer and third party access in the app and remove what you no longer need.
- Businesses with commercial systems should add monitoring devices and gateways to their asset inventory and patch them like any other internet facing system. The SolarView case involved a flaw that had been fixed for months.
How should grid equipment makers build secure products?
Grid equipment makers should develop to IEC 62443-4-1, sign firmware and verify it on the device, remove hard coded credentials and shared keys, support IEC 62351 and DNP3 Secure Authentication, ship an SBOM with every release, run a disclosure programme, and patch fast with clear mitigations.
Device design
- Follow IEC 62443-4-1 for secure development and design components to IEC 62443-4-2 requirements at the security level your customers need.
- Sign firmware and configuration packages, and verify signatures on the device before installing. Make remote firmware update something the utility can turn off.
- Remove hard coded credentials and shared keys. Use per-device credentials and keys provisioned securely at manufacture.
- Support IEC 62351 for MMS, GOOSE and Sampled Values, IEC 62351-5 or DNP3 Secure Authentication, and role based access control. Make them usable, with key management that a utility can actually operate.
- Keep protection functions isolated from communication functions so a communication fault cannot disable protection.
Engineering and cloud software
- Treat engineering software as security critical. The SEL research showed configuration import and local web services as attack paths.
- In clouds and APIs for meters, inverters and chargers, authorise every request against the specific object it touches, enforce TLS certificate validation in apps, and require approval and rate limits for fleet-wide setting changes.
- Protect signing keys and update infrastructure as the most sensitive assets you own.
Supply chain and disclosure
- Produce an SBOM for every firmware release, software build and image, and deliver it to customers.
- Track vulnerabilities in every component, including protocol stacks such as libiec61850 and cryptographic libraries, and tell customers which are reachable in your product.
- Run a coordinated vulnerability disclosure programme and publish advisories with affected versions and mitigations. Every case in this guide that ended well ended with a vendor fix after disclosure.
- Plan for long support lifetimes. Grid equipment stays in the field far longer than consumer electronics, and it needs security updates for all of that time.
How Safeguard helps teams that build and run grid software
Safeguard covers the software supply chain inside grid products in one platform: SBOMs for firmware, apps and images, binary and image scanning, vulnerabilities ranked by reachability, SCA, SAST and secrets scanning for code you own, fix pull requests gated by your CI, Gold patched packages, supplier risk and compliance evidence.
Safeguard does not sit on the substation LAN or in the protection scheme. It sits where most of the long-lived exposure in this guide comes from: the firmware, protocol stacks, open source components, engineering tools, apps and clouds that equipment makers build and utilities depend on. Teams that take this seriously often end up with separate tools for source scanning, binary analysis, SBOM management, supplier questionnaires and audit evidence. 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.
Figure 12. How Safeguard fits grid equipment and utility software, from inputs to fixes and evidence.
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 the grid that covers Windows engineering and configuration tools, SCADA and head-end installers, solar and charger apps on both mobile platforms, virtual appliances for control centre software, and the containers behind DER clouds. Firmware images for relays, RTUs, meters and inverters may need to be extracted into their file systems first, and the usual limits of binary analysis apply: encryption, heavy obfuscation or unusual packing can give incomplete results. Our guide to scanning app binaries with Safeguard and our explainer on binary composition analysis go further.
For a utility, this is a way to check what a vendor delivered before it is installed, and to answer "do we run an affected version of this library" across the whole estate.
Rank vulnerabilities by what can actually be reached
A gateway image or inverter firmware can contain hundreds of components, and many of their vulnerabilities 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 a maker, that is how you tell the MMS parser flaw that needs a release this month from an unused function in a bundled library. For a utility working through a CIP-007 patch evaluation, it is evidence for which patches to prioritise in the next outage window.
Scan the code you own
For relay and RTU application code, gateway and meter firmware, inverter and dongle code, engineering tools, apps and cloud APIs 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 signing keys, API tokens and hard coded credentials before they ship. Hard coded credentials appear again and again in the research above, from solar dongles to engineering software, and they are exactly what secrets scanning exists to catch. Insecure direct object references in fleet APIs, like those in the SUN:DOWN research, are the kind of authorisation flaw that code review and testing must catch before release.
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 asks an equipment maker for under a CIP-013 supply chain plan, what a maker's own customers ask for in IEC 62443 procurement, and what lets anyone answer "which products embed libiec61850, and which version" in minutes. Safeguard's Third-Party Risk Manager lets you request SBOMs from your own suppliers, score them and monitor them continuously, so the supplier side of the chain is visible too. See SBOMs for firmware and IoT devices and our firmware supply chain security guide.
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 requalification, Safeguard Gold provides patched, hardened builds of those packages, and Safeguard can fork and patch the vulnerable package so your next firmware build pulls a fixed version. For an equipment maker facing a long type test cycle, or a utility waiting on that maker, that is the difference between shipping the fix in the next planned release and waiting for the whole chain to move. Closed source components still have to come from their vendor, 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 teams working to IEC 62443-4-1 or supporting customers under NERC CIP-010 and CIP-013, the SBOMs, vulnerability monitoring records, risk decisions and fix trail are the raw material that assessors, customers and auditors ask for.
Deploy where your code lives
Relay source code, signing infrastructure and control system software 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
Grid security teams usually combine several kinds of tools, and they do different jobs. OT network monitoring platforms from companies such as Dragos, Nozomi Networks and Forescout watch live control system traffic, detect intrusions and support incident response in operational networks, which is work Safeguard does not do. Dragos has also said it acquired NetRise and runZero to extend its platform. Source SCA tools such as Snyk or Black Duck cover developers' code. GRC tools collect audit evidence. 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. It works alongside an OT monitoring platform, not instead of one. For more, see energy sector software security and NERC CIP and our explainer on IEC 61850, GOOSE and IEC 62351 security.
The takeaway
The grid has been hacked, and the details are less cinematic than the reputation. A macro in an email. A VPN that only wanted a password. Operator screens doing exactly what they were built to do. A firewall nobody knew was there. An inverter cloud that never checked whose plant it was returning.
The defences are not exotic either. Know what you run. Keep the business network and the vendors on the far side of a watched door. Watch the control traffic. Let protection act on its own. Keep known good firmware on a shelf and people who can switch a substation by hand. And because nearly every device in this story was built by someone else, out of components built by someone else again, make the software supply chain visible: SBOMs for every release, vulnerabilities ranked by what is reachable, and fixes that do not have to wait for every link in the chain to move at once.
Try it yourself
Click through these interactive Safeguard demos to see the steps described in this guide.
Test yourself
How did attackers cut power to about 225,000 customers in Ukraine in December 2015?
They entered through phishing months earlier, stole credentials, and used the operators' own SCADA screens to open breakers. They also overwrote substation gateway firmware and wiped computers, and crews restored power manually within hours.
Why did Dragos and SANS say opening breakers through the DMS is not a vulnerability?
They said it is a feature of the system. The real problem was that an adversary had access and operators had no way to detect the takeover.
What was Industroyer and why did it matter?
Industroyer was modular malware with payloads that spoke IEC 60870-5-101, IEC 60870-5-104, IEC 61850 and OPC DA, plus a wiper and a tool aimed at protection relays. It automated grid control using the protocols as designed, needing network access rather than protocol vulnerabilities.
How does vendor remote access put the grid at risk, and what closes most of the gap?
It is a deliberate path from the internet into operational networks, and attackers use stolen vendor credentials, exposed VPN gateways and unpatched firewalls. Jump hosts, multi-factor login, time limits and monitoring close most of the gap.
What is the difference between IEC 62443, IEC 62351 and NERC CIP?
IEC 62443 is a general standard family for securing industrial automation. IEC 62351 secures the power system protocols themselves, and NERC CIP is the mandatory set of reliability standards for the North American bulk electric system.
Where are most published solar inverter flaws found, and why does that matter?
Mostly in the vendor clouds, apps and dongles that manage many inverters at once, not in the power electronics. One flaw can reach thousands of sites, and researchers warn coordinated control of a large fleet could stress grid stability, though no such event has been publicly confirmed.
Frequently asked questions
Can hackers take down the power grid?
Attackers have caused real outages, most clearly in Ukraine in 2015 and 2016, but those were regional, lasted hours, and needed months of preparation. Grids have automatic protection, many independent operators and staff who can run substations by hand. A short local outage is possible for a prepared attacker; a long, wide one is far harder.
What happened in the Ukrainian power grid outages of 2015 and 2016?
In December 2015, attackers who entered through phishing used operators' own SCADA screens to open breakers at three distribution companies, cutting power to about 225,000 customers for several hours, and overwrote gateway firmware to slow recovery. In December 2016, an outage at a transmission substation near Kyiv lasted about an hour, and investigators later linked it to malware named Industroyer.
What was Industroyer and why did it matter?
Industroyer was modular malware with payloads that spoke IEC 60870-5-101, IEC 60870-5-104, IEC 61850 and OPC DA, plus a wiper and a tool aimed at protection relays. It mattered because it automated grid control using the protocols as designed, needing network access rather than protocol vulnerabilities.
Can smart meters be hacked?
Yes. The best documented real cases are fraud, where meters were reprogrammed through optical maintenance ports to under-record usage. Researchers have also shown flaws that could let attackers control meters over the utility network. The main design concern is the remote disconnect switch, which needs strong authentication and limits on mass commands at the head-end.
Can solar inverters be hacked, and could that affect the grid?
Yes. Research in 2024 and 2025, including Forescout's SUN:DOWN report with 46 vulnerabilities across Sungrow, Growatt and SMA products, found flaws mostly in vendor clouds, apps and dongles, which manage many inverters at once. Researchers warn that coordinated control of a large fleet could stress grid stability. No such grid event has been publicly confirmed, and the vendors fixed the reported issues.
What is the Purdue model and how does it relate to IEC 62443 zones and conduits?
The Purdue model divides industrial networks into levels, from physical equipment at level 0 to business systems at level 4, usually with a DMZ in between. IEC 62443 turns those levels into zones with target security levels, joined only by conduits that control and monitor traffic. Together they describe how to segment a utility network.
What is the difference between IEC 62443, IEC 62351 and NERC CIP?
IEC 62443 is a general standard family for securing industrial automation, covering asset owners, integrators and product makers. IEC 62351 secures the power system protocols themselves, such as IEC 61850, IEC 104 and DNP3. NERC CIP is the mandatory set of reliability standards for the North American bulk electric system, including supply chain rules in CIP-013.
How should utilities manage vendor remote access?
Route all vendor access through a jump host in an operations DMZ, require named accounts and multi-factor login, enable access per ticket for a limited time, record sessions, and disable it afterwards. Find and remove unknown remote paths such as contractor-installed routers or firewalls, and patch internet facing gateways first.
Sources
Every factual claim in this guide is linked inline to its source. The full list is in the front matter. Key references:
- E-ISAC and SANS ICS, Analysis of the Cyber Attack on the Ukrainian Power Grid (2016)
- Dragos, Ten Years After Ukraine's Power Grid Cyberattack
- ESET, Industroyer (2017) and Industroyer2 (2022)
- Forescout, SUN:DOWN press release (2025) and OT:ICEFALL (2022)
- Bitdefender, Solarman and Deye research (2024)
- DIVD, Enphase IQ Gateway case (2024)
- Nozomi Networks Labs, SEL RTAC vulnerabilities and SEL engineering software vulnerabilities (2023)
- Herath, Liu, Hong and Park, Real-time mitigation techniques for IEC 61850 and IEC 62351 substations (CIGRE 2025)
- Anderson and Fuloria, Who controls the off switch? (2010)
- Soltan, Mittal and Poor, BlackIoT (USENIX Security 2018)
- NERC, CIP standards CIP-005-7, CIP-007-6, CIP-010-4, CIP-013-2 and CIP-015-1