The 2018 Russian router alert needed no malware, so agency defenses came down to configuration and ownership
TA18-106A said Russian state actors took over routers through legacy management services and harvested credentials. For an agency, every fix depended on knowing who configured each device, its secrets and its software.
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
TA18-106A said Russian state actors took over routers through legacy management services and harvested credentials. For an agency, every fix depended on knowing who configured each device, its secrets and its software.
- The week On Monday, April 16, 2018, the Department of Homeland Security, the FBI and the United Kingdom’s National Cyber Security Centre published a joint Technical Alert, TA18-106A.
- Reachability: management services that answered anyone The alert described the first stage as reconnaissance.
- Secrets: the configuration file as a credential store The second stage in the alert was the extraction of configuration files.
- Integrity: knowing what the device actually ran The third part concerned what happened after the actors were inside.
- Ownership: devices the agency could not configure The fourth part is the one the alert addressed through contracts.
- Why an attack without malware changed what agencies measured Security programs measure what their tools can see.
- How the alert fell on different public organizations The same alert asked different things of different public organizations, depending on what they owned and who managed it.
- Objections: scale, cost, responsibility and relevance Four objections deserve a direct answer.
- What a government technology leader could check The four parts translate into checks that need no new tool, only an owner and an inventory.
- With hindsight The campaign the alert described did not end in 2018, and neither did the weaknesses.
The week
On Monday, April 16, 2018, the Department of Homeland Security, the FBI and the United Kingdom’s National Cyber Security Centre published a joint Technical Alert, TA18-106A. Its title was “Russian State-Sponsored Cyber Actors Targeting Network Infrastructure Devices.” It described the worldwide exploitation of routers, switches, firewalls and network-based intrusion detection devices. It said the targets were “primarily government and private-sector organizations, critical infrastructure providers, and the Internet service providers (ISPs) supporting these sectors.”
The alert said that since 2015 the U.S. government had received information from multiple sources that cyber actors were exploiting large numbers of enterprise-class and small office and home routers and switches worldwide. The U.S. government assessed that actors supported by the Russian government carried out the campaign. The alert said the FBI had high confidence that these actors were using compromised routers for man-in-the-middle attacks. It listed the purposes as supporting espionage, extracting intellectual property, maintaining persistent access and potentially laying “a foundation for future offensive operations.”
The alert’s central technical point was that the actors did not need new tools. It said they needed neither zero-day vulnerabilities nor installed malware to exploit these devices. Instead they took advantage of three conditions. These were devices with legacy unencrypted protocols or unauthenticated services, devices insufficiently hardened before installation, and devices no longer supported with security patches by their manufacturers.
It named the protocols. The actors scanned for Telnet, HTTP, the Simple Network Management Protocol (SNMP) and Cisco Smart Install, a management feature that listened on TCP port 4786. Specially crafted SNMP and Smart Install packets could make a device send its configuration file to an actor-controlled host using the Trivial File Transfer Protocol. The alert said the configuration file contained password hash values, from which actors could derive legitimate credentials, and SNMP community strings.
With those credentials, the alert said, “Legitimate user masquerade is the primary method” by which the actors exploited devices. They logged in as privileged users through Telnet, SSH or web management interfaces. They could extract more configuration data, export the operating system image, modify configurations, create Generic Routing Encapsulation (GRE) tunnels, or mirror or redirect traffic through infrastructure they controlled. The alert noted that a Smart Install Exploitation Tool had been posted to the internet on November 18, 2016.
The mitigation section addressed five audiences separately: all organizations, manufacturers, security vendors, ISPs, and owners or operators. It told all organizations not to allow internet access to the management interface of any network device and to disable Telnet and SNMP versions 1 and 2c. It told owners or operators to block TFTP leaving the organization for internet hosts and to verify that firmware came from a trusted source. It also gave them contract terms for the equipment their internet service provider fielded.
The alert was updated within the week. An update said that on April 19, 2018, an industry partner had notified the National Cybersecurity and Communications Integration Center and the FBI of activity matching the alert. The partner reported that actors had redirected DNS queries to their own infrastructure by creating GRE tunnels and had obtained configuration files of networked devices.
The same Monday, the National Institute of Standards and Technology released version 1.1 of its Framework for Improving Critical Infrastructure Cybersecurity. NIST said the update covered authentication and identity, self-assessing cybersecurity risk, managing cybersecurity within the supply chain, and vulnerability disclosure.
Alerts about state-sponsored hacking can read as threat intelligence for specialists. We read TA18-106A as a governance document for public agencies. Its key sentence was that the actors needed neither zero-day vulnerabilities nor malware. An attack that uses a device’s own management features with real credentials leaves little for malware scanners or endpoint tools to find. The defense therefore had to come from how devices were configured, how their secrets were kept, how their software was verified and who was responsible for them.
That mattered for agencies because network devices could be managed outside the security tooling that covered servers and desktops. The alert said that once installed, many network devices were not maintained at the same security level as desktops and servers. It added: “Owners and operators often overlook network devices when they investigate, examine for intruders, and restore general-purpose hosts after cyber intrusions.” For a federal agency, a state department or a county, the router at a field office could belong to the agency, a contractor or the carrier. Each answer meant a different path to fixing it.
We take the alert in four parts. Each part is a question an agency had to be able to answer for every network device it relied on. Together they cover the stages the alert described, from finding a device to controlling its traffic:
- Reachability: which management services the device exposed, and to whom.
- Secrets: what the device held that could unlock it or other devices, and where copies of that information traveled.
- Integrity: whether the device ran the software and configuration the agency believed it ran.
- Ownership: who had the authority and the obligation to fix the device, including devices the agency did not own.
Our argument is that the fourth question decided the other three. An agency could disable Telnet, protect configuration files and verify firmware only on devices it controlled. For devices in a carrier’s hands, its only lever was the contract, which the alert itself pointed to. After the four parts, we look at why the absence of malware changed what agencies needed to measure. We then cover how the alert fell on different public organizations, the main objections and what a government technology leader could check.
Reachability: management services that answered anyone
The alert described the first stage as reconnaissance. The actors scanned internet address space for enabled ports and services, fingerprinted devices and identified vulnerable ones. The protocols they looked for were Telnet, HTTP, SNMP on ports 161 and 162, and Smart Install on port 4786. Login banners and other data from enabled services could reveal a device’s make and model and information about its owner.
Every one of those services was a management feature. Each existed so that an administrator could reach the device remotely. The problem the alert described was that the services answered requests from anywhere, including from people who were not the administrator.
What the alert asked agencies to change
The general mitigations were specific. Organizations should not allow unencrypted management protocols such as Telnet to enter from the internet. They should not allow internet access to the management interface of any network device. Where remote management was needed, it should run through an encrypted VPN with mutual authentication, from an allowed network or host. Telnet, SNMPv1 and SNMPv2c should be disabled in favor of SSH and SNMPv3.
The alert went further on devices that could not comply. “DHS strongly advises owners and operators to retire and replace legacy devices that cannot be configured to use SNMP V3.” That sentence turned a configuration question into a procurement question. A device that could not be secured by settings had to be replaced.
Smart Install as a design problem
The alert singled out Smart Install. It described it as “an unauthenticated management protocol developed by Cisco” that allowed administrators to download or overwrite any file on a supporting router or switch. The feature existed to install and configure new devices remotely. Because it required no authentication, the alert said, its source address could be spoofed. Anyone who could reach port 4786 could use it.
The alert gave administrators commands to check whether Smart Install was enabled and signatures to detect its use. It also cautioned that the commands showed only whether the feature was on, not whether a device had been compromised. For an agency, this meant two separate tasks: turning off exposure and checking whether exposure had already been used.
Small office equipment at the edge
The alert addressed small office and home routers alongside enterprise equipment. It said few such devices ran antivirus or integrity tools, and that manufacturers shipped devices with exploitable services enabled for ease of installation. For agencies, that class of device appeared at small field offices, pop-up sites and, where permitted, staff homes.
An agency could not manage every home router. It could decide which kinds of sites were allowed to connect to agency systems through which paths. A field office running a consumer-grade router with remote management enabled was an agency device in effect, whoever bought it. Deciding in advance what equipment such sites could use, and who configured it, kept those devices inside the inventory rather than outside it.
What an agency needed to know about reachability
To act on this part, an agency needed a list of its network devices. For each, it needed to know which management services were enabled and from where they could be reached. That list was harder to produce than a server inventory. Network devices could be added by field staff or contractors, and their management settings changed outside the change process. An external scan of the agency’s own address space would show what an attacker would see. An internal review of device configurations would show what was enabled but filtered.
In our reading, the alert’s guidance amounted to a rule that management should happen only on a management network. The rule was simple to state and harder to apply across hundreds of field sites. Each site where a device was managed directly over the internet was an exception that someone had to own.
Exposure the agency could not see from inside
An internal review shows what a device is configured to do. It does not always show what the internet can reach. Address ranges change, temporary access rules outlive their purpose and devices at remote sites can sit outside central firewalls. The alert’s reconnaissance stage worked from the outside in. An agency checking its own exposure needed the same view.
A recurring external scan of the agency’s own address space would list every management service that answered. Each result could then be matched to a device record and an owner. A result with no matching record was itself a finding: a device the agency did not know it had, or one that a contractor or carrier had placed without telling it.
Secrets: the configuration file as a credential store
The second stage in the alert was the extraction of configuration files. Crafted SNMP or Smart Install packets could make a device send its configuration to an outside host over TFTP. If a network blocked external SNMP at the boundary, the actors spoofed source addresses to appear internal. The resulting files gave the actors what they needed for the next step.
What a configuration file contained
The alert said a configuration file “contains a significant amount of information about the scanned device, including password hash values.” Those values allowed actors to derive legitimate credentials. The file also held SNMP community strings and other network information. That let the actors build maps of the network and plan further exploitation.
That made the configuration file a secret in its own right. It was routinely copied for backup, troubleshooting and audit. The alert noted that actors could extract password hashes “from configurations sent by owners and operators across the Internet.” Each copy was a potential source of credentials.
Passwords that unlocked more than one device
The alert’s exploitation stage depended on credentials. In some cases the actors used brute force against Telnet and SSH. “However, for the most part,” the alert said, actors obtained legitimate credentials easily. Default passwords, weak password policies and passwords derivable from harvested data all helped. The alert recommended changing default passwords immediately, giving each device a unique password and, where possible, using key-based two-factor authentication instead of passwords.
Unique passwords mattered because of the configuration file. If one file yielded a hash for a password shared across the agency’s devices, every device with that password was exposed. An agency that managed hundreds of routers with a common administrative credential had, in effect, one secret protecting all of them.
Community strings as passwords
The alert listed SNMP community strings among the contents of a configuration file. A community string works as a shared password for SNMP access. With SNMP versions 1 and 2c, it travels across the network unencrypted. An actor holding the string could query a device and, depending on how access was set, change it.
That made community strings part of the same credential problem. An agency that set one community string across its network, or kept a default, had a single secret whose exposure opened every device that accepted it. The alert’s advice to move to SNMPv3, which supports authentication and encryption, addressed this directly. Until then, unique strings limited to trusted management hosts reduced the damage a leaked file could do.
Configuration backups
Backups of device configurations were a necessity for recovery and a risk for exposure. A backup server holding every router’s configuration held every router’s password hashes and community strings. It deserved the protection given to a password vault. Access should be limited to the people who restore devices, and its logs should be reviewed.
Where configuration copies traveled
The alert recommended blocking TFTP from leaving the organization for internet hosts. It said devices should send configuration data to a secured host on a trusted segment of the internal management network. That addressed one path. The others were human: configurations attached to support tickets, shared with vendors or stored on file shares. An agency could not protect a secret whose copies it had not counted.
For an agency, this part implied a records question. Where were device configurations stored, who could read them, and when were they sent outside the agency? A configuration repository with access controls and logging would answer it. A habit of emailing configurations to a vendor’s support desk would not. Rotating device credentials after any suspected exposure of a configuration file was the remedy the alert’s logic pointed to.
Integrity: knowing what the device actually ran
The third part concerned what happened after the actors were inside. The alert described installation and command and control. Through Smart Install, actors could overwrite files, modify configurations or upload maliciously modified operating system or firmware images. Logged in as privileged users, they could export the operating system image, create GRE tunnels and mirror or redirect traffic.
Why a modified device was hard to see
The alert said network devices had writeable file systems where malware for other platforms could be stored. It noted that few network devices, especially small office and home models, ran antivirus or integrity tools. A device altered by a privileged user looked, from outside, like a device configured by its administrator. The only reliable comparison was against a known-good configuration and a verified software image.
What the alert asked for
The owner and operator mitigations included verifying that firmware and operating systems came from a trusted source and were issued by the manufacturer. The alert pointed to vendor guidance for validating images. Its detection guidance included inspecting protocol 47 traffic to or from unexpected addresses and unexplained creation, modification or destruction of GRE tunnels in logs. The April 19 update gave that guidance immediate weight, since the industry partner had reported DNS redirection through GRE tunnels.
Why redirected traffic mattered for public services
The alert said actors could mirror or redirect traffic through infrastructure they controlled, and the April 19 update described DNS queries redirected through GRE tunnels. For a public agency, redirected traffic could include citizen requests to agency websites, connections between offices and calls to other agencies’ systems. Traffic protected by strong encryption from end to end would reveal less to an actor in the middle. Traffic sent in clear text, including legacy management protocols, would reveal everything.
That gave agencies a second reason to remove unencrypted protocols. Disabling Telnet and old SNMP versions closed doors into devices. Encrypting traffic between sites limited what an actor on a compromised device could read. The alert addressed the first directly. The second followed from its description of what control of a router allowed.
Baselines as the control
For an agency, integrity meant three records. The first was a stored, approved configuration for each device. The second was a record of the software image and how it had been verified. The third was a log of changes made outside the change process. Comparing the running configuration with the approved one would surface added tunnels, changed routes or new accounts. Comparing image hashes with the manufacturer’s published values would surface tampering.
The three records required no new security products. They required that network operations treat configurations as controlled documents. Where a network team kept configurations for its own use, the alert, in our reading, made them part of the security record.
Restoring a device after suspected compromise
The alert said owners and operators overlooked network devices when restoring hosts after intrusions. That gap had a practical cause. Incident response plans could describe how to rebuild a server and reset user accounts while leaving the network path untouched. If an actor held a router, a clean server would sit behind a compromised device.
A restoration step for network devices followed from the alert’s guidance. It meant reloading a verified software image and re-applying the approved configuration. It meant rotating device credentials and community strings, and reviewing logs for tunnel creation and unexpected management sessions. The step needed an owner in the incident plan, and that owner needed access to the baselines described above.
Ownership: devices the agency could not configure
The fourth part is the one the alert addressed through contracts. Its mitigations for ISPs said they should not field equipment in the network core or at customer premises with legacy, unencrypted or unauthenticated services. ISPs should plan to upgrade unsupported equipment and should patch fielded equipment or tell customers how to. Among the factors that made devices vulnerable, the alert listed that ISPs “do not replace equipment on a customer’s property when that equipment is no longer supported by the manufacturer or vendor.”
What the alert told owners to put in contracts
The owner and operator section turned those expectations into contract terms. Owners should “Specify in contracts that the ISP providing service will only field currently supported network equipment and will replace equipment when it falls into an unsupported state.” They should also specify that the ISP would regularly apply updates and patches, or notify customers and give them the ability to apply them.
For an agency, those two sentences were the only mitigation in the alert it could apply to a router it did not own. The device at a field office provided by a carrier, or the edge router supplied with a circuit, sat outside the agency’s configuration authority. The contract defined what the carrier owed. If the contract was silent on supported equipment and patching, the agency had no basis to require either.
What a contract review would look for
Reviewing a carrier or managed service contract against the alert was a short exercise. The reviewer would look for a clause requiring supported equipment, a clause on patching or notice of patches, and a clause on replacement when equipment reached end of support. The reviewer would also look for terms on how management access to the device was restricted and who held its credentials.
Where those clauses were missing, the agency had two options. It could amend the contract at the next opportunity, such as renewal or a task order. Or it could ask the provider for a written statement of its practices and attach it to the contract file. Neither option fixed a device on its own. Both gave the agency a basis to ask for evidence and to escalate when the answer was unsatisfactory.
End-of-life devices as an ownership question
The alert listed devices no longer supported with security patches as one of three conditions the actors exploited. Whether such a device stayed on the network was a decision someone had to make and fund. For an agency-owned device, that was a budget and refresh question. For a carrier-owned device, it was a contract question. For a contractor-managed device, it depended on whether the managed service contract covered lifecycle replacement.
Contractor-managed networks
Between the agency and the carrier sat a third case: networks run by a contractor under a managed service agreement. The contractor configured the devices, held the credentials and kept the configurations. Each of the first three parts of the alert then depended on the contractor’s practices. The agency’s lever was again the contract and the evidence it required, such as configuration baselines, credential rotation records and lists of devices nearing end of support.
Evidence is the practical test of a managed service relationship. An agency could ask its contractor for a monthly list of devices with their software versions and support dates. It could ask for the date each device’s administrative credentials were last changed. It could also ask for any configuration change made outside an approved change request. If the contractor could not produce those, the agency learned something about its exposure, whatever the contract said.
In our reading, the ownership question explains why the alert addressed manufacturers, security vendors and ISPs as well as owners. Manufacturers decided defaults. ISPs decided what equipment they fielded. Owners decided configuration. Security vendors decided what detection existed. An agency was an owner for some devices and a customer for others, and it needed to know which was which before it could act.
Why an attack without malware changed what agencies measured
Security programs measure what their tools can see. A program measured by endpoint coverage, malware detections and server patch compliance would barely register the campaign the alert described. No malware was required. No zero-day was used. The actors logged in.
Measures that would have shown the exposure
The measures that fit the alert were different. They included the share of network devices with management interfaces reachable from the internet. They included the share still running Telnet, SNMPv1 or SNMPv2c, the share with Smart Install enabled and the share using unique credentials. Others were the number past end of support and the number whose running configuration matched an approved baseline.
Each of those measures depended on a complete inventory of network devices. That was the first gap the alert exposed. An agency that could not list its routers and switches could not report on any of them.
Why detection could not carry the load
The alert gave detection guidance: signatures for Smart Install traffic, log review for Telnet sessions, correlation of inbound SNMP with outbound TFTP. Those were useful, and they depended on the agency collecting logs and flow data from network devices. Where monitoring centered on hosts, that data might not reach the security team. The alert’s guidance would only help where network device telemetry already did.
Even with telemetry, detection came after exposure. The alert’s own structure placed prevention first: disable the services, protect the secrets, verify the images, and contract for the devices the agency did not own. Detection was the backstop for the cases where those steps had not been taken.
How the alert related to the framework released the same day
NIST’s Cybersecurity Framework version 1.1 appeared on the same day. NIST said the update addressed authentication and identity, self-assessment, supply chain risk and vulnerability disclosure. An agency mapping the alert to the framework would find its demands falling under asset management, access control, configuration management and supplier relationships. The coincidence of dates was just that. But it gave agencies a ready structure for turning the alert into program measures.
Reporting the measures upward
Measures that no one sees change nothing. An agency that adopted the measures above would need to decide who received them. For a federal agency, that could be the chief information security officer and the authorizing officials for the systems the network served. For a state or county, it could be the chief information officer and the official who signs network contracts.
The measures also had a natural order. Inventory completeness came first, because every other measure depended on it. Exposure and credential measures came next, because they described what an outside actor could reach. Baseline and lifecycle measures came last, because they described how quickly the agency could detect change and remove weak devices.
What auditors and inspectors could ask
Agency auditors and inspectors general review security programs against stated policies and standards. The alert gave them concrete questions. Did the agency know how many network devices it had? Could it show which ones exposed management services? An agency that could answer those from records was in a different position from one that had to run a project to find out.
We would add one more line of inquiry: whether network devices appeared in the agency’s incident response and continuity plans. The alert’s point that owners overlooked network devices during restoration suggested that plans could omit them. A plan that named the network device owner and the restoration steps closed that gap on paper, which is where an auditor would look first.
How the alert fell on different public organizations
The same alert asked different things of different public organizations, depending on what they owned and who managed it. We look at four kinds of organization separately.
Federal civilian agencies
Federal agencies could operate large, distributed networks with field offices connected through carrier services. They also had a central authority for directives. In April 2018, DHS could issue binding operational directives to federal civilian agencies. The alert itself was guidance rather than a directive, and acting on it depended on each agency’s own program.
For a federal agency, the practical first step was a device inventory that distinguished agency-owned, contractor-managed and carrier-owned equipment. The second was a contract review for the carrier and managed service agreements, measured against the alert’s two contract sentences.
A federal agency also had a place to record the decisions the alert forced. System security plans and authorization packages describe the controls on the systems a network supports. A network device past end of support, or one managed by a carrier without a patching clause, is a risk that an authorizing official can accept, reject or require to be fixed. Recording that decision made the exposure visible to the official accountable for it.
State agencies and statewide networks
State governments could run statewide networks that served several agencies, sometimes through a central IT department and sometimes through a carrier contract. In that model, one contract could govern devices across hundreds of sites. Adding the alert’s supported-equipment and patching terms to that one contract would reach all of them at once.
Counties, cities and special districts
Smaller local governments may rely on a carrier’s equipment or on a single managed service provider. They may have few staff to review device configurations. For them, the ownership part was the whole problem. One available step was to ask the provider, in writing, which devices it managed, which were past end of support, and whether Telnet, SNMP versions 1 and 2c and Smart Install were disabled.
Local governments may also buy network services through cooperative purchasing agreements or state contracts. Where that was the route, the terms the alert recommended would have to appear in the master agreement to reach the local buyer. A county could still add them through its own order or statement of work, if the master agreement allowed.
Public utilities and other operators of control systems
The alert gave one example that applied directly to public utilities. An actor controlling a router between industrial control system sensors and controllers, it said, could manipulate messages and create configurations that could lead to loss of service or physical destruction. For a municipal water or power utility, network devices in the control environment carried that consequence. The reachability and integrity parts applied with particular force there.
Objections: scale, cost, responsibility and relevance
Four objections deserve a direct answer. Each has force, and none, in our reading, changes the argument that ownership determined what an agency could do.
That agencies had too many devices to review
The first objection is scale. A large agency could have thousands of network devices across hundreds of sites. Reviewing each configuration by hand was impractical. We agree, and that is why the inventory and baseline came first. Automated collection of running configurations, compared against approved baselines, scales in a way manual review does not. The alert’s checks, such as whether Telnet or Smart Install was enabled, could be expressed as rules applied to every collected configuration.
That replacement costs too much
The second objection is cost. Replacing devices that cannot run SNMPv3 or receive patches requires capital that agencies may not have budgeted. The alert did not ignore this. It framed replacement as the remedy for devices that could not be secured by settings. Where replacement had to wait, the other parts still applied: restrict reachability to a management network, rotate credentials and watch for configuration changes. An agency that documented its unsupported devices and the compensating steps had a defensible plan.
That the carrier was responsible
The third objection is that carrier-owned devices were the carrier’s problem. Legally, the carrier’s obligations depended on the contract. Operationally, the agency’s traffic passed through those devices whatever the contract said. The alert’s phrase “Own the Router, Own the Traffic” made the point. An agency could not transfer the consequence of a compromised edge router by pointing to the carrier. It could only transfer the obligation to prevent it, and only in writing.
That an intelligence campaign was beyond an agency’s concern
A fourth objection is relevance. A small agency might assume that a campaign attributed to a foreign government targeted larger organizations. The alert answered this in its own text. It named government organizations among the primary targets and described the scanning as broad-scale as well as targeted. A device found by a broad scan needs no particular reason to be chosen. In our reading, the conditions the alert described were what made a device a target, more than the identity of its owner.
What a government technology leader could check
The four parts translate into checks that need no new tool, only an owner and an inventory. They are ways, not legal requirements, to see whether the conditions the alert described exist in an agency’s network today.
- Reachability: a current list of network devices whose management interfaces can be reached from the internet, with an owner and a plan for each.
- Secrets: where device configurations are stored, who can read them, whether each device has a unique administrative credential, and when credentials were last rotated.
- Integrity: whether each device has an approved configuration baseline and a verified software image, and how differences are reported.
- Ownership: for each device, whether the agency, a contractor or a carrier configures it, and whether that party’s contract requires supported equipment and patching.
- Lifecycle: a list of network devices past or nearing end of support, with the date and funding source for replacing each.
Each check produces a record. Together they show that the agency treated its network devices as systems with owners, secrets and baselines, as the alert asked, rather than as plumbing.
With hindsight
The campaign the alert described did not end in 2018, and neither did the weaknesses. On April 18, 2023, the UK NCSC, NSA, CISA and FBI published a joint advisory, AA23-108. It described exploitation of Cisco routers in 2021 by a group the authors called APT28 and assessed to be almost certainly a unit of Russia’s military intelligence service. The advisory said the group had used SNMP access into Cisco routers worldwide. It said this included a small number in Europe, U.S. government institutions and approximately 250 Ukrainian victims.
The 2023 advisory named the same weaknesses as the 2018 alert. It said weak SNMP community strings, including the default “public,” allowed the actors to gain access to router information. The compromised routers accepted SNMP v2 requests, which send community strings unencrypted. For some devices the actors exploited a vulnerability that Cisco had announced on June 29, 2017, with patched software available. On unpatched devices, the advisory said, they deployed malware. The 2018 alert had told owners to disable SNMPv1 and v2c and to retire devices that could not run SNMPv3.
Federal civilian executive branch agencies then received a binding requirement on the reachability question. CISA’s Binding Operational Directive 23-02, issued June 13, 2023, applied to routers, switches, firewalls, VPN concentrators, proxies, load balancers and out-of-band server management interfaces. It covered devices whose management interfaces used protocols such as HTTP, SNMP, Telnet, TFTP and SSH over the public internet. It applied to systems hosted by agencies “or third parties on their behalf.” Within 14 days of notification or discovery, agencies had to remove such an interface from the internet or protect it with a separate policy enforcement point as part of a zero trust architecture.
The directive’s background explained why. It said threat actors had adjusted to better endpoint detection “by targeting network devices supporting the underlying network infrastructure.” In our reading, that was the 2018 alert’s point about attacks without malware, now stated as federal policy.
On December 4, 2024, CISA, NSA, the FBI and partners in Australia, Canada and New Zealand published guidance on visibility and hardening for communications infrastructure. They warned that actors affiliated with the People’s Republic of China had compromised networks of major global telecommunications providers. The guidance said identified exploitations aligned “with existing weaknesses associated with victim infrastructure; no novel activity has been observed.” Its first monitoring advice was to investigate configuration changes to network devices outside the change management process.
On August 20, 2025, the FBI warned that cyber actors it attributed to the Russian Federal Security Service’s Center 16 were exploiting SNMP and end-of-life networking devices. Those devices were running an unpatched Smart Install vulnerability, CVE-2018-0171. The FBI said that in the past year it had detected the actors collecting configuration files for thousands of networking devices associated with U.S. entities across critical infrastructure sectors. On some devices, they modified configuration files to enable unauthorized access. The FBI pointed readers back to the 2018 Technical Alert as guidance that “remains relevant.”
The ownership and lifecycle question also reached directive form. CISA’s Binding Operational Directive 26-02, issued February 5, 2026, addresses end-of-support edge devices, including routers and firewalls. It requires federal civilian agencies to inventory devices on a CISA list of end-of-support edge devices and to decommission them on stated schedules. It also requires a process for continuous discovery of edge devices and an inventory of those that are, or will soon be, out of support. Its required actions cover systems owned or operated by agencies or on their behalf.
What held up from 2018 was the diagnosis. Legacy management protocols, configuration files carrying credentials, and unsupported devices were still the conditions named in advisories in 2023, 2024 and 2025. What changed was the instrument. For federal civilian agencies, guidance became binding directives on management interfaces and end-of-support devices. State and local agencies are outside those directives. In our view, the 2018 alert’s contract sentences remain their main lever over devices they do not configure themselves.
Spotted an error? Tell us and we will check it and correct the article where needed.