CrashOverride issued valid grid commands; the monitoring rules that followed stopped at the bulk power system

On June 12, 2017, DHS warned that CrashOverride sent control systems’ own unauthenticated commands. Catching that meant watching inside control networks. FERC’s 2023 and 2025 orders brought such monitoring to the bulk power system only, leaving public operators of local systems to act on advice.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

On June 12, 2017, DHS warned that CrashOverride sent control systems’ own unauthenticated commands. Catching that meant watching inside control networks. FERC’s 2023 and 2025 orders brought such monitoring to the bulk power system only, leaving public operators of local systems to act on advice.

  1. The week On June 12, 2017, the Department of Homeland Security’s National Cybersecurity and Communications Integration Center published an alert on malware called CrashOverride.
  2. The commands: an attack that spoke the system’s language The alert’s first listed capability was the one that set CrashOverride apart.
  3. The visibility: seeing an attack inside the control network If the commands look legitimate, detection has to look elsewhere.
  4. The recovery: restoring control when the tools were the target The third part covers what happens after an attack lands.
  5. The reach: who was bound to act The fourth part concerns obligation.
  6. What it meant for organizations that depend on public systems Banks, hospitals and agencies that do not run control systems still depend on them.
  7. Objections, in order The alert said IEC 101, IEC 104 and IEC 61850 were more commonly used outside the United States in electric power control systems. [1] It also said the general tactics could be modified to affect US infrastructure, and that similar malware could reach beyond electric power. [1] CISA’s 2022 advisory described tools that interacted with devices through normal management protocols and Modbus. [6] The approach traveled.
  8. What an agency’s control system owner could check The four parts translate into checks any public operator can run, whether or not a federal standard applies to it.
  9. With hindsight The alert itself changed.

The week

On June 12, 2017, the Department of Homeland Security’s National Cybersecurity and Communications Integration Center published an alert on malware called CrashOverride. Its alert code was TA17-163A. [1] The alert listed the systems affected as industrial control systems. [1]

The alert began with its sources. The center said it was aware of public reports from two security firms, ESET and Dragos. Those reports described a new, highly capable attack platform for industrial control systems. [1] According to the reports, the platform was used in 2016 against critical infrastructure in Ukraine. [1] The center said it was working with partners to validate the firms’ analysis and to understand the risk to US critical infrastructure. [1]

The alert rated the activity under the center’s incident scoring system as Yellow, or medium priority. [1] It stated there was no evidence the malware had affected US critical infrastructure. It added that the tactics described could be modified to target US networks and systems. [1]

The technical analysis described a scalable platform. The reported modules focused on organizations using the industrial protocols IEC 101, IEC 104 and IEC 61850. [1] The alert noted these protocols were more commonly used outside the United States in electric power control systems. [1] It said the platform fundamentally abused the functionality of the target’s legitimate control system to achieve its intended effect. [1]

The alert listed five reported capabilities. The malware issued valid commands directly to remote terminal units over industrial protocols. [1] One command sequence, as reported by Dragos, toggled circuit breakers in a rapid open-close-open-close pattern. [1] The alert said this could cause individual utilities to island from infected parties and degrade grid reliability. [1]

The other capabilities followed. The malware could deny service to local serial ports on Windows devices, cutting off communications with field equipment. [1] It could scan and map a control environment using several protocols, including OPC. [1] It could exploit a denial-of-service flaw in a Siemens relay, leaving the relay to be reset by hand. [1] It included a wiper module that rendered Windows systems inert, requiring a rebuild or restoration from backup. [1]

On detection, the alert was frank. It described CrashOverride as second-stage malware able to operate independently of its initial command and control. [1] Traditional methods of detection might therefore not find an infection before the malware ran. [1] The center encouraged behavioral analysis to identify activity that came before it. It published indicators of compromise and detection signatures, several of them credited to Dragos. [1]

The mitigations ran long. They included application whitelisting on static systems such as human-machine interfaces. [1] They called for authentication and authorization so that field devices verify commands before acting, and ICS firewalls where legacy hardware could not. [1] They asked operators to keep separate credentials for corporate and control networks, and never to share directory services between them. [1]

The alert also asked for backups of gold images, engineering workstation images and PLC and RTU configurations, with offline copies of install media. [1] It recommended segmenting networks into enclaves and restricting remote access. It asked operators to monitor at four places, including inside the control network itself. [1] And it recommended a restoration plan with known-good images ready. [1]

The CrashOverride alert published file hashes and detection signatures for the malware itself. [1] Its more important passage described something those tools would not catch. According to the reports it relied on, the malware sent valid commands, in the control system’s own protocols, to field equipment that did not check who was asking. [1] A breaker that opens on command cannot tell a dispatcher from an intruder.

Our argument is that this design moved the decisive defense inside the control network. The alert’s own advice pointed there: authenticate commands, watch traffic within the control LAN, and keep known-good copies to restore. [1] Federal reliability rules later moved monitoring inside control networks: a 2023 order directed NERC to write the requirement, and a 2025 order approved the standard. [2] [3] Those rules reach the bulk power system. The statute excludes facilities used in local distribution. [3] Public agencies that run local power, water and building systems were left to act on advice.

We take the alert in four parts:

  1. The commands: what it means when an attack uses the system’s own instructions.
  2. The visibility: what it takes to see such an attack inside a control network.
  3. The recovery: how to restore control when the attack targets the tools of control.
  4. The reach: which organizations were bound to act, and which were advised.

The first part covers the attack itself. The second covers detection. The third covers restoration. The fourth covers the rules and their boundaries. Together they span an incident from the first command to the return of normal operations, and the obligations around it. After the four parts, we look at public organizations that depend on these systems, answer objections in order, and list what an agency’s control system owner could check.

A note on sources and our reading. We describe the malware only as the alert described it, citing the firms it relied on. [1] We describe what followed from the Justice Department, CISA and the Federal Energy Regulatory Commission. [2] [3] [4] [5] [6] [7] Statements about what a rule requires of a particular entity are our reading. Utilities and agencies should confirm their obligations with counsel and their regulators.

The commands: an attack that spoke the system’s language

The alert’s first listed capability was the one that set CrashOverride apart. It issued valid commands directly to remote terminal units over industrial control protocols. [1] The commands were not malformed and did not exploit a coding flaw. They were the instructions an operator would send.

Protocols built for trust

The alert said the malware exploited the lack of authentication and authorization in common control protocols to issue unauthorized commands to field devices. [1] In such a protocol, a device acts on a correctly formed message from a station it expects. It does not ask for proof of identity.

In our reading, that design made sense for closed serial networks, where only the control center could reach a device. It makes less sense once the control network shares routes, workstations or credentials with the business network. The alert warned against that sharing. It warned against shared directory services and trust stores between corporate and control zones. [1]

The breaker pattern

The alert gave one concrete example of a command sequence. As reported by Dragos, the malware could toggle circuit breakers in a rapid open-close-open-close pattern. [1] The alert said this could create conditions where individual utilities island from infected parties, degrading grid reliability. [1]

The example matters for its shape. Each individual command was legitimate. The harm lay in the sequence and the speed. A defense that checks each message alone would approve every one. A defense that knows how breakers are normally operated, by whom and at what pace, could see the pattern as abnormal.

Mapping before acting

The alert also said the malware scanned and mapped control environments using several protocols, including OPC, and that mapping improved the payload’s chance of success. [1] OPC is a common way for software to read and write values on control equipment. A tool that maps through it learns the names and addresses of the devices it will later command.

In our reading, mapping is a promising moment for a defender. It happens before harm, and it produces traffic that differs from daily operations. A human-machine interface polls the same points every few seconds. A scan asks about everything once. The difference is visible to anyone recording the traffic.

Why signature thinking falls short

Signatures find known code. The alert did publish signatures and hashes, to find CrashOverride’s own files. [1] But the commands that toggled breakers would look the same whether sent by malware or by a dispatcher. A tool that inspects only for malicious code would pass them.

The alert said as much. It warned that the general tactics could be reused with modified technical implementations against US infrastructure. [1] In our reading, what endured was the commands, not the binaries. The next attack could use new code to send the same kind of valid instruction.

Authentication where the device allows it

The alert’s first answer was to make field devices verify commands before acting. [1] Newer equipment and newer protocol versions can support that. Some older equipment cannot, and the alert said so: in some instances, legacy hardware may not be capable. [1]

For legacy gear, the alert offered two options. Asset owners could place ICS firewalls in front of devices to inspect and authenticate commands statefully. Or they could upgrade the field devices. [1] Both cost money and engineering time. Both need an inventory of which devices speak which protocols.

Credentials as the real perimeter

The alert also turned to people’s accounts. It warned that adversaries increasingly sought legitimate credentials, especially privileged ones, because those let them act like real users and leave less evidence. [1] It called for multifactor authentication where possible, least privilege and separate credentials for corporate and control zones. [1]

That advice connects the protocol problem to identity. If the field device trusts any station that can reach it, the control is who can reach that station. Engineering workstations and human-machine interfaces become the gate. Their accounts, remote access paths and software become the perimeter.

Buying devices that can say no

Authentication has to be designed in or bought in. A public operator replaces field devices on long cycles, through public procurement. Each replacement is a chance to require that the device can verify commands, log them and accept signed firmware. The alert asked operators to insist that vendors digitally sign updates, or publish hashes through a separate channel. [1]

Specifications are cheap to write and expensive to omit. A device bought without command authentication will stay in service for its working life. The decision made in a procurement office therefore sets the security of a substation for years. In our reading, the alert gave procurement officers language they could cite in a solicitation.

What this part asks of a public operator

For a city electric department or a water authority, the commands problem translates into three tasks. The first is an inventory of field devices, their protocols and whether they can authenticate commands. The second is a plan for the devices that cannot. The third is a review of every account and path that can reach a station able to send commands. [1]

None of these tasks is new to control engineers. The alert reframed them as security work with owners and deadlines. A public operator would also need procurement language, so that new devices arrive able to verify commands.

The visibility: seeing an attack inside the control network

If the commands look legitimate, detection has to look elsewhere. The alert pointed defenders toward behavior and toward the network inside the control zone. [1] This second part is about what that requires.

Second stage, no beacon

The alert described CrashOverride as a second-stage capability. It could operate independently of its initial command and control. [1] The alert warned that traditional detection methods might not find an infection before the malware ran. [1]

The alert therefore asked organizations to look for precursor activity: the reconnaissance, credential use and movement that comes before the payload. [1] It noted that the malware scanned and mapped control environments using several protocols. [1] Mapping traffic on a control network is itself a signal, if someone knows what normal mapping looks like.

Four places to watch

The alert recommended monitoring programs in four places. They were the internet boundary, the boundary between business systems and the control demilitarized zone, the boundary between that zone and the control LAN, and inside the control LAN. [1] In our reading, the fourth place is the one that bears on valid commands. [1]

The alert asked operators to watch traffic on ICS boundaries for abnormal communications and to monitor traffic within the control network for malicious connections or content. [1] It asked them to analyze logins by time and place and to watch account administration. [1] Each of those steps needs a sensor, a log and a person who reviews it.

Remote access as a watched path

The alert treated remote access as a path to be narrowed and watched. It asked operators to remove remote accesses wherever possible, especially modems. [1] It asked that any remaining access be operator controlled, time limited and procedurally similar to lock out, tag out. [1] It asked that vendors and employees use the same paths, with no double standards, and that remote persistent vendor connections not be allowed. [1]

Each of those rules creates a record. A time-limited session has a start, an end and an approver. A single path for vendors and staff means one place to log. Where possible, the alert preferred monitoring-only access enforced by data diodes over read-only access enforced by software settings. [1] For a public operator, that list doubles as a design for where to put sensors.

Baselines before alarms

A monitor inside a control network can only flag what differs from normal. That requires a baseline: which stations talk to which devices, using which protocols and commands, at what times. FERC later wrote the same idea into its directive, asking for a baseline of network activity built from traffic and data flows. [3]

Control networks suit baselines. Their traffic is regular, and the set of legitimate commands is small. A rapid open-close-open-close pattern on breakers, as the alert described, should stand out against a baseline. [1] It stands out only if someone has recorded the baseline and set the rule.

Evidence that survives the attack

Detection also needs records an attacker cannot erase. The alert’s wiper module rendered Windows systems inert. [1] Logs kept on those systems would go with them. FERC’s later directive asked for logs of sufficient fidelity to investigate an incident, with measures to keep attackers from removing evidence. [3]

For an operator, that means sending control network logs to a separate store, with its own access controls. It means keeping enough of the traffic to reconstruct what commands were sent and from where. It also means deciding how long to keep them, a choice with cost.

Where the sensors go in a small utility

A large control center can afford sensors at every boundary. A small public utility has to choose. The alert’s four places give an order of priority. [1] In our reading, a small operator would start at the boundary between business systems and the control zone, where intruders cross. It would add a sensor inside the control LAN next, where valid commands travel.

Passive monitoring suits control networks because it does not send traffic of its own. A sensor that listens on a mirrored port cannot trip a breaker. That reassures engineers who worry, with reason, that active scanning can disturb fragile devices. The choice of passive tools is a safety decision as well as a security one.

What the indicators were good for

The alert published indicators of compromise in two formats, and a set of detection rules. [1] Those had real value in June 2017. An operator could search its systems for the specific files the firms had found. A clean result was reassuring about that toolkit.

It was not reassuring about the technique. The alert itself said the sources provided did not constitute an exhaustive list. [1] A modified version would carry new hashes. In our reading, the indicators answered whether an operator had this malware, and the behavioral advice answered whether it would see the next one.

Who watches

The hardest part of visibility is staffing. A sensor on a control LAN generates alerts. Someone who understands both the protocol and the process must decide whether a command was normal. In a large utility, that may be a dedicated team. In a small public utility, it may be the same engineers who run the system.

The alert did not settle that question, and could not. It described what to monitor. [1] Who monitors, and with what training, is a budget and workforce decision for each operator. In our reading, that decision is where advice to a small public utility meets its limit.

The recovery: restoring control when the tools were the target

The third part covers what happens after an attack lands. CrashOverride, as the alert described it, attacked the means of recovery as well as the means of control. [1] Its wiper, its serial port denial and its relay exploit each made restoration harder.

The wiper and the rebuild

The alert said the wiper rendered Windows systems inert, requiring a rebuild or backup restoration. [1] In a control center, those Windows systems include human-machine interfaces, historians and engineering workstations. Without them, operators lose their screens and engineers lose the tools to reconfigure devices.

The alert listed what to keep. It named server gold images, ICS workstation gold configurations, engineering workstation images, PLC and RTU configurations, passwords and configuration information, and offline install media. [1] Each item answers a specific step in a rebuild.

Ports and relays that need hands

Two capabilities pushed recovery into the field. The malware could deny service to serial ports on Windows devices, preventing communications with field equipment over serial links. [1] And it could exploit a Siemens relay flaw that left the relay needing a manual reset. [1]

A relay that must be reset by hand needs a technician at the substation. In our reading, that turns a cyber incident into a field operations event. It needs crews, vehicles, access to sites and a plan for which sites to visit first.

Manual operation as the fallback

When screens are dark and devices cannot be trusted, operators fall back to manual control. The alert did not describe manual operations in detail. It did recommend a restoration plan with gold disks ready to restore systems to known good states. [1] A restoration plan that assumes working screens would fail in exactly the scenario the alert described.

A sound plan, in our reading, would answer three questions. Which functions can run by hand, and for how long. Which staff can run them. And in what order systems come back once clean images are ready. Those answers belong to operations as much as to security.

Credentials after an incident

The alert’s response plan was blunt about credentials. It said a plan might include disconnecting internet connections, disabling affected accounts, isolating suspect systems and immediately resetting 100 percent of passwords. [1] In a control environment, that is a large job. Shared accounts on old devices, service accounts and vendor accounts all need new secrets.

An operator that has never inventoried those accounts cannot reset them all. The reset list is therefore a preparation task. It should name every credential, where it is stored and who can change it, before the incident that requires the change.

Clean software for the rebuild

A rebuild needs software that can be trusted. The alert asked operators to obtain updates from authenticated vendor sites, validate downloads and never load updates from unverified sources. [1] It asked that updates go first to a test system with malware detection. [1] During recovery, the same rules apply under time pressure. Offline copies of install media, which the alert listed, spare the operator from downloading under attack. [1]

Testing the restore

A backup that has never been restored is a hope. The alert asked operators to keep gold images and configurations. [1] CISA’s later advisory on control system tools added hashing and integrity checks on firmware and controller configuration files, to confirm the backups were valid. [6]

For a public operator, a restore test needs a test bench or a maintenance window. It needs the engineering workstation software, licenses and project files the gold images depend on. It needs the people who know how to load a relay configuration. Those dependencies are where tests find gaps.

Exercises with operations staff

A restoration plan is tested by people as much as by software. A tabletop exercise that pairs control room operators with security staff will find gaps a document review misses. Someone has to decide to isolate the control network, someone has to authorize manual operation, and someone has to call the vendor for clean media. The alert’s response plan included escalation triggers, investigation and public affairs. [1] Each of those needs a named person.

Paying for recovery capacity

Recovery capacity is spare capacity. It sits unused until the bad day. Public budgets find spare capacity hard to defend. In our reading, the alert’s list gives a budget office something concrete: a set of named backups, a restore test schedule and a field response plan, each with a cost.

The reach: who was bound to act

The fourth part concerns obligation. The June 2017 alert was advice. It told critical infrastructure companies to follow best practices and pointed them to a joint DHS, NSA and FBI paper on defending control systems. [1] Whether any organization had to follow it depended on other law.

An alert binds no one

The alert used the language of recommendation: operators should, operators are encouraged, NCCIC recommends. [1] It did not cite a rule or a deadline. For a utility outside any mandatory regime, the alert was information to weigh against other priorities.

Advice has strengths. It can move faster than a rule, and it can describe a threat while the facts are still being validated, as this alert did. [1] Its weakness is that it competes for money with work that is required. A public utility facing an audit on another matter will fund the audit first. In our reading, that competition for money is one likely reason voluntary guidance would be adopted unevenly by small operators.

The bulk power system

The bulk power system had a mandatory regime. The Federal Power Act gives FERC authority to approve reliability standards developed by NERC, the electric reliability organization. [3] FERC’s later orders describe the critical infrastructure protection standards as part of that regime. [2] [3] [4]

The Act defines the bulk-power system as facilities and control systems needed to operate the interconnected transmission network, and generation needed for reliability. It excludes facilities used in the local distribution of electric energy. [3] In our reading, that boundary placed a distribution substation outside the federal standards, however critical it was to the town it served. The line is not a clean split between utility types. FERC’s October 2026 notice on the updated monitoring standard lists distribution providers among the respondents to its information collection. [7] In our reading, an entity that runs distribution can still be registered for equipment that supports the bulk system, and the boundary falls on the equipment, not on the owner.

Public bodies inside and outside the line

Some public bodies sit inside the federal regime. Commenters on FERC’s 2023 directive included the Bonneville Power Administration and the California Department of Water Resources State Water Project. [2] The American Public Power Association and the Large Public Power Council joined trade association comments on that directive and later ones. [2] [3] [4]

Others sit outside it. A city that runs only distribution, a water utility or a transit agency would not fall under the electric reliability standards for those systems. For them, the protections the alert described would come from their own policies, their state’s rules or their funding conditions. We do not attempt to map those here.

Public operators as commenters

Public operators are regulated parties in this regime too. The Bonneville Power Administration commented on FERC’s 2023 monitoring directive. It asked FERC to limit the directive to high impact systems, citing resources and timing. [2] FERC declined. [2]

That exchange shows the tension for a public operator. A public operator faces the same budget limits as any other, and the same attacker. FERC’s answer was to keep medium impact systems with external connectivity in scope. [2] Federal and state operators outside the regime face the same question without a regulator to answer it for them.

Impact categories inside the regime

Even inside the federal regime, obligations vary by impact category. FERC’s 2023 directive covered high impact systems and medium impact systems with external routable connectivity. [2] It asked NERC to study the low impact systems and the medium impact systems without such connectivity. [2] A small registered entity with only low impact systems faced different duties from a large control center.

The impact categories are a reasonable way to rank federal attention. They also mean that the inside-the-network monitoring the alert recommended in 2017 arrived where FERC judged the impact highest. [1] [2]

The federal standard as a template

For operators outside the regime, FERC’s later work offers a ready template. The three objectives it set for monitoring are a baseline of normal traffic, detection of unauthorized activity, connections, devices, protocols and software, and logs protected from tampering. [3] Those objectives are written without reference to bulk power equipment. In our reading, a water authority or a city electric department could adopt them by policy.

Adopting a template is not the same as being examined against it. A registered entity faces audits and penalties. A public operator that adopts the same objectives voluntarily answers to its own board or council. The difference is accountability, not content.

Funding and procurement

Where mandates stop, money and contracts carry the weight. A public operator can write command authentication, logging and restore testing into equipment and integration contracts. It can make them conditions of its capital plan. The alert gave such an operator a specification it could cite. [1] Whether it did so was a local choice.

What it meant for organizations that depend on public systems

Banks, hospitals and agencies that do not run control systems still depend on them. A breaker that opens on a forged command cuts power to a data center as surely as to a substation. The alert described the potential to degrade grid reliability. [1]

Hospitals

A hospital depends on its utility and on its own building systems. Its generators and building automation run on control systems too. The alert’s advice on separate credentials, segmentation and gold images applies to those systems. [1] In our reading, a hospital’s facilities team holds a version of the same problem the alert described.

The organizational question is who owns it. Building systems can be bought and run by facilities staff, outside the information security program. The alert’s advice assumes a single owner for credentials, segmentation and backups across a control environment. [1] A hospital that splits those duties between facilities, clinical engineering and IT needs a written agreement on who does each. The same is true of an agency that shares a building with tenants or contractors who run the mechanical plant.

Banks

A bank’s exposure is mostly indirect. Its data centers, branches and payment operations depend on power and telecommunications. Business continuity plans that assume short outages would deserve a second look after an alert describing manual relay resets. [1] For a bank, the alert speaks to the duration of outages more than their cause. A plan built for a storm assumes crews can restore power once roads clear. A plan for a cyber incident has to allow for crews who must first confirm that each device can be trusted.

Agencies that own buildings and fleets

Federal, state and local agencies run building management systems, traffic signals and fleet depots. Some of those systems use the same kinds of protocols and workstations the alert described. The alert’s monitoring and recovery advice reaches them, without any rule requiring it. [1]

Mapping the dependency

A dependent organization cannot secure its utility’s control network. It can learn how its utility would restore service after a cyber incident. The alert’s recovery advice gives it questions to ask: whether gold images and device configurations are kept offline, and whether field crews can reset relays by hand. [1] The answers shape how long backup power must last.

For a hospital or a county emergency operations center, that conversation belongs in continuity planning. It sets the fuel contracts for generators, the order in which services are shed, and the triggers for moving operations. A cyber incident at a utility looks, from outside, like any long outage. The difference is that restoration may be slower and less predictable.

Objections, in order

1. The protocols named were mostly used abroad

The alert said IEC 101, IEC 104 and IEC 61850 were more commonly used outside the United States in electric power control systems. [1] It also said the general tactics could be modified to affect US infrastructure, and that similar malware could reach beyond electric power. [1] CISA’s 2022 advisory described tools that interacted with devices through normal management protocols and Modbus. [6] The approach traveled.

2. Authenticating commands is impossible on legacy equipment

The alert agreed that some legacy hardware could not authenticate commands. [1] It offered ICS firewalls as the alternative. [1] Where neither is possible, monitoring inside the network becomes the main control. That supports our argument that the decisive defense moved inside.

3. Air gaps make internal monitoring unnecessary

The alert itself assumed paths into control networks. It warned against remote persistent vendor connections, infected laptops and shared trust stores. [1] FERC’s 2023 directive noted that attackers inside a perimeter can perform the same functions as an authorized user, as the Ukraine attacks showed. [2] A gap that holds is valuable. Monitoring confirms whether it holds.

4. Small public utilities cannot afford this

Cost is the strongest objection. A small utility may have no security staff for its control network. The alert’s list is long. [1] Our answer is that the list can be ranked. Separate credentials, a device inventory and tested restores cost less than full monitoring. They also make later monitoring possible.

5. The federal rules addressed the problem

FERC did act, through Orders 887, 907 and 918. [2] [3] [4] Those orders apply within the reliability regime, which excludes local distribution. [3] For public bodies outside the regime, the rules show what federal regulators thought was needed. They do not require it of them.

6. Attribution matters more than technique

The government did not attribute the activity in the June 2017 alert. It attributed it in the 2021 update. [1] Attribution matters to diplomacy and prosecution, and the Justice Department’s 2020 charges followed that path. [5] For an operator deciding what to build, the technique matters more. A different actor can send the same valid commands.

7. Monitoring inside the network produces noise no one can read

Noise is a fair concern, and staffing is the real constraint. Control networks, though, carry fewer kinds of traffic than business networks, on more regular schedules. A baseline of legitimate stations and commands is short. [3] Departures from it are fewer and more meaningful than alerts on a corporate network. The harder task is finding people who can judge them.

What an agency’s control system owner could check

The four parts translate into checks any public operator can run, whether or not a federal standard applies to it.

  1. Which field devices accept commands without authentication, and what stands in front of each one?
  2. Do corporate and control networks use separate credentials and separate directory services?
  3. Is there a recorded baseline of normal traffic inside the control network, and does anyone review departures from it?
  4. When was a gold image, a relay configuration and a controller program last restored from backup, and by whom?
  5. If a substation relay needed a manual reset tonight, who would go, and how long would it take to reach every site?

The first two checks address the commands. The third addresses visibility. The fourth and fifth address recovery. The reach question sits behind all five: if no rule requires them, the agency’s own policy has to. [1] [3]

With hindsight

The alert itself changed. CISA, which now hosts the alert, revised it on July 20, 2021. [1] The update said the US Government attributed the activity to Russian nation-state cyber actors. It assessed that those actors deployed CrashOverride in a cyberattack against Ukrainian critical infrastructure. [1]

Criminal charges had come earlier. On October 19, 2020, the Justice Department announced that a federal grand jury in Pittsburgh had returned an indictment on October 15, 2020. [5] It charged six people it described as officers in Unit 74455 of Russia’s military intelligence directorate. [5] The charges are allegations, and the defendants are presumed innocent unless proven guilty. [5]

According to the department, the alleged conduct included destructive malware attacks from December 2015 through December 2016 against Ukraine’s electric power grid, Ministry of Finance and State Treasury Service. [5] The department named the malware as BlackEnergy, Industroyer and KillDisk. It said KillDisk and Industroyer each caused blackouts in Ukraine. [5] The same indictment alleged the NotPetya attacks of June 27, 2017. [5]

The technique did not stay with one tool. On April 13, 2022, the Energy Department, CISA, NSA and FBI warned of custom tools built to target industrial control devices. [6] The tools could scan for, compromise and control certain Schneider Electric and OMRON controllers and OPC UA servers. [6] One tool’s modules interacted with devices through normal management protocols and Modbus. [6] The advisory recommended continuous monitoring of internal systems and communications for hostile actions and lateral movement. [6]

The federal reliability rules moved toward the alert’s fourth monitoring place. FERC issued Order No. 887 on January 19, 2023, published February 9, 2023. [2] It directed NERC to develop standards requiring internal network security monitoring for all high impact systems and medium impact systems with external routable connectivity. [2] It noted that attackers inside a security perimeter could perform the same functions as an authorized user, as the Ukraine attacks demonstrated. [2]

NERC filed CIP-015-1 on June 24, 2024. FERC approved it in Order No. 907, issued June 26, 2025, and published July 2, 2025. [3] The standard addressed three objectives FERC had set: a baseline of network activity, detection of unauthorized activity, and logs of enough fidelity, protected from tampering. [3] FERC found it applied only within the electronic security perimeter and directed NERC to extend it to access control and monitoring systems outside that perimeter. [3]

NERC answered with CIP-015-2, filed June 18, 2026, which extends monitoring to those systems. [7] A FERC notice of October 13, 2026, said the change was implemented through an order in Docket RD26-6-000 issued August 10, 2026. [7] For low impact systems, FERC approved CIP-003-11 in Order No. 918, issued March 19, 2026. [4] It added controls to authenticate remote users, protect authentication data in transit and detect malicious communications at low impact assets with external routable connectivity. [4] FERC estimated 1,673 affected entities, of which 406 were small. [4]

What held up from June 2017? The alert’s diagnosis held: attacks would use the control system’s own functions, and detection had to happen inside the network. [1] [2] [6] The federal rules eventually required much of what the alert advised, for the bulk power system. [2] [3] [4] The statute’s exclusion of local distribution did not change. [3] For public operators outside the regime, the 2017 alert’s list remains advice, adopted by choice, contract or state rule.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • CrashOverride
  • industrial control systems
  • public utilities
  • NERC CIP
  • network monitoring