INTELLIGENCE
ZERO|TOLERANCE
Intelligence Advisory
CRITICAL PRIMARY SOURCE

APT28 GRU Router Hijacking Turns Home Wi-Fi Into Enterprise Attack Surface

Apr 7, 2026 · CVE-2023-50224 · 23 TP-Link models · 182 malicious IPs · CRITICAL

Publication Date
2026-04-07
Category
Nation-State & Espionage
Author
K. Ellabban
Organization
Zero|Tolerance Security Research

On April 7, 2026 the UK National Cyber Security Centre published an advisory titled "APT28 exploit routers to enable DNS hijacking operations." The same day, the FBI's Internet Crime Complaint Center issued alert I-040726-PSA, "Russian GRU Exploiting Vulnerable Routers to Steal Sensitive Information," joined by the US National Security Agency and partners from Canada, the Czech Republic, Denmark, Estonia, Finland, Germany, Italy, Latvia, Lithuania, Norway, Poland, Portugal, Romania, Slovakia and Ukraine.

Both documents describe the same operation. Russian military intelligence has been compromising small office and home office routers, rewriting the DNS server addresses those routers hand out over DHCP, and using the resulting control of name resolution to run adversary-in-the-middle attacks against Microsoft Outlook sessions. The harvest is passwords and OAuth tokens. The entry point named by both governments is CVE-2023-50224, a vulnerability disclosed in 2023 affecting a TP-Link router line that no longer receives support.

Executive Summary

KEY FACTS

  • WhatCompromise of SOHO routers to overwrite DHCP/DNS settings, redirect selected lookups to actor-controlled resolvers, and conduct adversary-in-the-middle interception of the resulting sessions.
  • WhoNCSC does not quantify victims and characterises the activity as opportunistic. Microsoft Threat Intelligence separately counted more than 200 organizations and 5,000 consumer devices affected. Microsoft names government, information technology, telecommunications and energy sectors, including at least three government organizations in Africa. NCSC notes interactive operations against a small number of MikroTik routers, often located in Ukraine.
  • HowExploitation of internet-reachable router management interfaces, likely via CVE-2023-50224 on TP-Link hardware, to extract stored credentials and then submit a second crafted HTTP GET request altering the router's DHCP DNS configuration.
  • DataPasswords, OAuth or similar authentication tokens, emails, and web browsing information normally protected by TLS.
  • ActorAPT28. NCSC states, "We assess that APT28 is almost certainly the Russian General Staff Main Intelligence Directorate (GRU) 85th Main Special Service Centre (GTsSS) Military Intelligence Unit 26165."
  • ImpactCredential theft, data manipulation and broader compromise, per NCSC. Because the interception point sits on the employee's home network, the compromise leaves no artefact on any enterprise-managed asset.
Incident Overview

WHAT HAPPENED

NCSC dates the activity to "Since 2024 and into 2026," during which APT28 configured virtual private servers to operate as malicious DNS servers. Those servers received high volumes of DNS requests originating from routers the actor had already exploited. Investigators sorted the infrastructure into two banner-pattern clusters.

In the first cluster, the DHCP DNS server settings of compromised SOHO routers were modified to include actor-owned IP addresses. Those settings were then inherited automatically by every downstream device on the network, including laptops and phones. This is the mechanism that makes the technique effective: nothing is installed on the endpoint, and the endpoint's own configuration is never touched. It simply asks the router where to send DNS queries and is told to send them to Russian military intelligence.

The resolvers did not poison everything. NCSC states that lookups for domain names containing key terms associated with particular services, often email applications or login pages, were resolved to further actor-owned IP addresses, while DNS requests not matching the actor's targeting criteria were resolved to the legitimate IP addresses for the requested services. Selective poisoning of this kind is a deliberate evasion choice. A household whose banking, streaming and social media all work normally has no reason to suspect its router, and a helpdesk ticket is never raised.

The redirected connections were then subjected to adversary-in-the-middle attacks. NCSC records that this activity could be conducted against both user browser sessions and desktop applications, and that harvested authentication material could include both passwords and OAuth or similar authentication tokens. The FBI describes the interception as effective against encrypted traffic where users navigate through a certificate error warning, and Microsoft states that its infrastructure presented an invalid TLS certificate that victims had to click past.

Follow-on logins with the stolen material, NCSC warns, may originate from infrastructure not listed in the advisory.

NCSC assesses the operation as opportunistic, with the actor gaining visibility of a large pool of candidate target users and then filtering down at each stage of the exploitation chain to triage for victims of likely intelligence value. This is bulk collection with a funnel attached, not targeted intrusion.

The infrastructure did not survive disclosure untouched. The FBI states that the Department of Justice and the FBI recently disrupted a GRU network of compromised small office home office routers used to facilitate malicious DNS hijacking operations. The PSA gives neither the date nor the scope of that disruption, and neither advisory says what proportion of the 182 published addresses it removed, so defenders should treat the indicator list as live rather than historical.

The FBI credits Black Lotus Labs at Lumen and Microsoft Threat Intelligence with technical contributions to the announcement.

Attribution

THREAT ACTOR

NCSC assesses that APT28 is almost certainly the Russian General Staff Main Intelligence Directorate (GRU) 85th Main Special Service Centre (GTsSS) Military Intelligence Unit 26165, and describes it as a highly skilled threat actor. The word "almost certainly" is the assessment NCSC published; it should not be restated as a confirmed fact.

The aliases NCSC lists are Forest Blizzard, Fancy Bear, STRONTIUM, the Sednit Gang and Sofacy. The FBI uses APT28, Fancy Bear and Forest Blizzard, and identifies the unit as the GRU 85th Main Special Service Center. Microsoft tracks the actor as Forest Blizzard and attributes this specific activity to a subgroup it tracks as Storm-2754.

NCSC records prior attributions to the same actor, including cyber attacks against the German parliament in 2015 involving data theft and disruption of email accounts belonging to German Members of Parliament and the Vice Chancellor, and an attempted attack against the Organisation for the Prohibition of Chemical Weapons in April 2018 aimed at disrupting independent analysis of chemical weapons used in the UK. NCSC also cross-references its earlier advisory on APT28 exploiting a known vulnerability to conduct reconnaissance and deploy malware on Cisco routers, which places this campaign in a documented multi-year pattern of edge-device targeting rather than treating it as a novel departure.

Paul Chichester, NCSC Director of Operations, said: "This activity demonstrates how exploited vulnerabilities in widely used network devices can be leveraged by sophisticated hostile actors. We strongly encourage organisations and network defenders to familiarise themselves with the techniques described in the advisory and to follow the mitigation advice."

Impact Assessment

WHAT WAS EXPOSED

  • Account passwords for web and email services, captured through adversary-in-the-middle interception of browser sessions and desktop mail applications.
  • OAuth or similar authentication tokens. These matter more than the passwords. A stolen token can be replayed to reach a mailbox without triggering a password prompt and, depending on configuration, without a second factor challenge, because the token already represents a completed authentication.
  • Email content and web browsing information normally protected by SSL and TLS, per the FBI.
  • Microsoft 365 and Outlook Web Access session material, based on the five Outlook domains NCSC lists as targeted for redirection.
  • Router administrative credentials, disclosed by CVE-2023-50224 before any DNS change was made.
  • DNS query metadata for every device behind a compromised router. Even where no interception followed, the actor-controlled resolver saw the full lookup history of the household, which is a standing intelligence collection capability against the individual.
Root Cause Analysis

TECHNICAL FAILURE CHAIN

1
End-of-support hardware left on the internet. The TP-Link models named are consumer kit from a generation that no longer receives firmware updates.

CVE-2023-50224 has a patch path that does not exist for most owners. The FBI's first recommendation is to upgrade end-of-support devices, which is an acknowledgement that patching is not available.

2
Router management interfaces reachable from the internet. The flaw is in the httpd service on TCP port 80.

NCSC maps the initial access to T1190, Exploit Public-Facing Application, and its lead mitigation is that management interfaces must never be exposed to the internet. Remote management enabled by default or by an ISP provisioning script converts a local flaw into a global one.

3
Stored credentials retrievable without authentication. CWE-290 in the router's web service meant the administrative password could be read by an unauthenticated request.

No brute force, no phishing, no user interaction.

4
DHCP as an unauthenticated trust channel. Client devices accept whatever DNS server the router advertises.

There is no mechanism on a standard laptop or phone to challenge that instruction, so a single router change silently reconfigures every device on the network.

5
Plaintext DNS with no validation. Neither DNSSEC validation nor encrypted transport (DoH or DoT) was in force on the affected paths.

Had endpoints resolved over an enforced encrypted resolver, the router's DHCP-supplied address would have been irrelevant.

6
TLS certificate warnings treated as an obstacle rather than a stop condition. Both the FBI and Microsoft state that the interception only succeeded against encrypted traffic where the user clicked through an invalid certificate warning.

The final control in the chain was a human decision, and it failed often enough to matter.

7
Bearer tokens without binding. OAuth tokens captured in transit remained usable because they were not cryptographically bound to the device or client that obtained them.

Token protection and continuous access evaluation exist specifically to break this replay path and were evidently not in force.

8
No enterprise visibility into the interception point. The compromised asset belongs to an employee, sits outside the corporate network, appears on no asset inventory, and is covered by no patch cycle or EDR agent.

The enterprise's first possible detection is an anomalous sign-in, which is several steps downstream of the actual compromise.

Detection

INDICATORS OF COMPROMISE

VULNERABILITY
  • CVE-2023-50224TP-Link TL-WR841N dropbearpwd improper authentication information disclosure - CVSS v3.0 6.5 (ZDI as CNA), AV:A/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N - CWE-290 - public advisory ZDI-23-1808 (internal candidate ID ZDI-CAN-19899) - CISA KEV since September 3, 2025
VPS BANNER PATTERNS (NCSC)
  • Banner pattern 1: SSH on TCP port 56777, plus "dnsmasq-2.85" on UDP port 53
  • Banner pattern 2: SSH on TCP port 35681, plus "dnsmasq-2.85" on UDP port 53 (DNS software present on only some servers in this pattern)
TARGETED DOMAINS REDIRECTED TO AITM INFRASTRUCTURE (NCSC)
  • autodiscover-s.outlook[.]com
  • imap-mail.outlook[.]com
  • outlook.live[.]com
  • outlook.office[.]com
  • outlook.office365[.]com
  • NCSC notes further non-Outlook domains were also observed but does not list them
INFRASTRUCTURE
  • NCSC published 182 unique IPv4 addresses across two clusters: 149 in cluster one (malicious DNS servers and AitM infrastructure) and 33 in cluster two (DNS hijacking and wider router operations).
  • Cluster one netblocks5.226.137[.]0/24, 23.106.120[.]0/24, 37.221.64[.]0/24, 64.120.31[.]0/24, 77.83.197[.]0/24, 79.141.160[.]0/24, 79.141.161[.]0/24, 79.141.173[.]0/24, 185.117.88[.]0/24, 185.117.89[.]0/24, 185.237.166[.]0/24
  • Cluster two netblocks64.44.154[.]0/24, 77.83.198[.]0/24, 79.141.173[.]0/24, 79.143.87[.]0/24, 88.80.148[.]0/24, 89.150.40[.]0/24, 103.140.186[.]0/24, 185.234.73[.]0/24
  • Representative addresses5.226.137[.]151, 37.221.64[.]77, 77.83.197[.]37, 79.141.161[.]66, 185.117.88[.]22, 185.237.166[.]55, 64.44.154[.]227, 103.140.186[.]148, 185.234.73[.]58. The complete list is in the NCSC advisory; NCSC cautions that specific selectors are liable to change and recommends holistic detection of DNS hijacking and AitM activity rather than reliance on this list.
TARGETED HARDWARE (NCSC LISTS 23 TP-LINK MODELS AND STATES, "IT IS LIKELY THAT THIS LIST IS NOT EXHAUSTIVE")
  • RoutersWR841N, WR841N/WR841ND, WR841HP, WR840N, WR842N, WR842ND, WR845N, WR941ND, WR945N, WR740N, WR740N/WR741ND, WR749N, WR1043ND, WR1045ND, MR6400, MR3420, Archer C5, Archer C7, WDR3600, WDR4300, WDR3500
  • Access pointsWA801ND, WA901ND
  • Do not treat these 23 as a closed set. Absence of a model from this list is not evidence it was not targeted.
  • MikroTik devices were also compromised; NCSC names no MikroTik models and no MikroTik CVE.
ACTOR TRACKING NAMES
  • APT28, Forest Blizzard, Fancy Bear, STRONTIUM, the Sednit Gang, Sofacy (NCSC); Storm-2754 (Microsoft subgroup designation)
MITRE ATT&CK
  • T1190Exploit Public-Facing Application
  • T1557Adversary-in-the-Middle
  • T1583.002Acquire Infrastructure: DNS Server
  • T1583.003Acquire Infrastructure: Virtual Private Server
  • T1584.008Compromise Infrastructure: Network Devices
  • T1586Compromise Accounts
  • T1588.006Obtain Capabilities: Vulnerabilities
Compliance Impact

REGULATORY EXPOSURE

  • UK GDPR / Data Protection Act 2018 - Articles 5(1)(f), 32, 33 and 34. Where an employee's mailbox was accessed via a hijacked home router, the employer remains the controller. Article 32 requires security appropriate to the risk taking account of the state of the art. Personal data accessed through the mailbox triggers Article 33 notification to the ICO within 72 hours and Article 34 notification to affected individuals where the risk is high. The penalty tiers differ, and the distinction is the one most often got wrong: under DPA 2018 section 157(6), infringements of Articles 32, 33 and 34 attract the standard maximum of GBP 8.7 million or 2 percent of total annual worldwide turnover, whichever is higher. Only a breach of the Article 5(1)(f) integrity and confidentiality principle reaches the higher maximum under section 157(5), GBP 17.5 million or 4 percent.
  • EU GDPR - Articles 5(1)(f), 32, 33 and 34, engaged for any affected controller established in the EU or processing EU residents' data. Article 83(4) places infringements of "Articles 8, 11, 25 to 39 and 42 and 43" in the lower tier at up to EUR 10 million or 2 percent of total worldwide annual turnover, whichever is higher. Articles 32, 33 and 34 all sit inside that range, so the security and notification failures this campaign exposes are 2 percent exposure, not 4. Only the basic principles under Article 83(5), which covers Articles 5, 6, 7 and 9, reach EUR 20 million or 4 percent, and for this incident that means Article 5(1)(f) alone. The remote-work configuration is squarely in scope: Article 32 obligations do not stop at the office perimeter, and a controller that permits mailbox access from unmanaged home networks without compensating controls owns that risk.
  • NIS2 Directive - Essential and important entities in the energy, telecommunications, digital infrastructure and public administration sectors are named by Microsoft among those affected. NIS2 imposes a 24-hour early warning and 72-hour incident notification obligation, and holds management bodies personally accountable for approving and overseeing cybersecurity risk measures.
  • CISA Binding Operational Directive 22-01 - CVE-2023-50224 entered the KEV catalog on September 3, 2025 with a remediation due date of September 24, 2025. US federal civilian executive branch agencies were under a binding obligation to remediate or discontinue use of affected products more than six months before these advisories published. Exploitation continuing into 2026 indicates the underlying device population was never addressed outside the federal estate, where the directive has no reach.
  • SEC cybersecurity disclosure rules - US-listed issuers must report material cybersecurity incidents on Form 8-K Item 1.05 within four business days of a materiality determination. Compromise of executive mailboxes through this vector is a plausible materiality trigger, and the fact that the initial access occurred on personal equipment does not alter the disclosure analysis.
  • FTC Act Section 5 - Reasonableness of security practices for organizations that permitted corporate email access from unmanaged devices, and for device vendors whose representations about product security do not match their end-of-support behaviour.
  • US state breach notification statutes - Triggered where mailbox contents accessed by the actor contained residents' personal information. Notification duties attach to the employer, not the employee whose router was compromised.
Analytical Limitations

INTELLIGENCE GAPS

?
The scale figures do not come from either government.

NCSC declines to quantify victims and calls the activity opportunistic; the FBI gives no numbers. The widely repeated "over 200 organizations and 5,000 consumer devices" is Microsoft Threat Intelligence's count, reported in its coverage of activity since at least August 2025. Microsoft does not explicitly tie the count to that window, so it should not be read as a figure for the full campaign NCSC dates to 2024 and into 2026. The two should not be merged.

?
The CVE link is hedged in the primary source.

NCSC says APT28 exploited the WR841N "likely using CVE-2023-50224." The FBI states it without qualification. No public artefact reconciles the difference, and NCSC names no CVE at all for the other 22 TP-Link models or for any MikroTik device.

?
CVE-2023-50224 is scored AV:A, adjacent network, by the Zero Day Initiative, yet NCSC maps initial access to T1190, Exploit Public-Facing Application.

Either the flaw is reachable from the internet on affected firmware in a way the CVSS vector does not capture, or a different access path preceded exploitation. The FBI's mitigation advice tells device owners to disable remote management interfaces from the Internet, which supports the reading that WAN-exposed management is what converted an adjacent-network flaw into a remotely exploitable one, but no source states this directly. Neither advisory resolves the discrepancy, and NIST has not published its own score for the CVE.

?
How the MikroTik devices were compromised is unstated.

NCSC describes interactive operations against MikroTik routers in Ukraine but names no vulnerability, no credential source and no models, so defenders with MikroTik estates cannot determine their exposure from this advisory.

?
Neither government states how many harvested credentials or tokens were successfully replayed.

NCSC notes only that subsequent malicious logins may originate from infrastructure not listed, which means published IP blocklists will not catch the account-takeover stage.

?
No router-resident malware, implant or persistence mechanism is described.

It is not stated whether a factory reset clears the compromise or whether the actor re-exploits reset devices, though NCSC's observation that some routers had both DNS entries poisoned indicates repeat exploitation of the same hardware.

?
The scope of the law enforcement disruption is unstated.

The FBI confirms that the Department of Justice and the FBI recently disrupted a GRU network of compromised SOHO routers used to facilitate malicious DNS hijacking operations, but gives no date, no device count and no indication of how much of the published infrastructure was seized. Separately, The Hacker News reports the operation under the DoJ codename Operation Masquerade and the campaign name FrostArmada, which it attributes to Black Lotus Labs at Lumen, with a peak of more than 18,000 unique IP addresses across at least 120 countries in December 2025. The FBI does credit Black Lotus Labs at Lumen and Microsoft Threat Intelligence with technical contributions to its announcement, which identifies the research behind those figures, but the DOJ press release returned an authorization error when retrieved for this report and the Black Lotus Labs publication was not obtained. Those codenames and figures therefore remain unverified secondary claims and are excluded from the analysis above.

Assessment

ZERO|TOLERANCE Advisory

1
Enforce encrypted DNS on the endpoint and stop trusting DHCP. Configure managed laptops to use a corporate DNS-over-HTTPS or DNS-over-TLS resolver by policy, and block the endpoint from falling back to DHCP-supplied resolvers.

On Windows estates, Microsoft's Zero Trust DNS enforcement is the specific control; it prevents the endpoint from reaching any destination whose name was not resolved by an approved resolver. This single change makes a poisoned home router irrelevant, because the router's DNS advertisement is never honoured.

2
Move Microsoft 365 to phishing-resistant authentication and bind the tokens. Deploy FIDO2 security keys or passkeys in Microsoft Authenticator for all administrative and high-risk accounts, then enable token protection and continuous access evaluation so a token lifted in transit cannot be replayed from attacker infrastructure.

Restrict MFA and passkey registration to trusted locations and compliant devices, so a stolen session cannot be used to enrol a new authenticator.

3
Require the corporate VPN or Zero Trust client for all mail access, with no exceptions for webmail. The FBI's own guidance to organizations is to review policies on how employees access sensitive data, including VPN use.

The interception in this campaign happens on the home LAN; an always-on tunnel established before name resolution removes the home network from the trust path entirely. Blocking Outlook on the web from unmanaged devices via Conditional Access closes the browser route that the five targeted domains depend on.

4
Treat employee home routers as in-scope assets and fund their replacement. The FBI recommends organizations consider incentivizing employees to upgrade outdated personal devices used for remote access.

Make it concrete: inventory what remote staff connect through, publish a supported-hardware list, and pay for the replacement of end-of-support kit. A GBP 60 router is cheaper than an Article 33 notification, and no amount of employee diligence patches a device the vendor has abandoned.

5
Disable WAN-side management on every edge device and verify it externally. NCSC's lead mitigation is that management interfaces must never be exposed to the internet.

Verify from outside the network rather than trusting the device's own settings page, since the flaw exploited here lives in the router's web service on TCP port 80.

6
Hunt for the specific artefacts now, not the IPs alone. Query outbound UDP/53 and TCP/443 against the netblocks NCSC published, but also alert on any endpoint using a resolver outside your approved set, on TLS certificate validation failures for outlook.office365[.]com and the other four listed domains, and on successful sign-ins carrying a high aggregate risk score.

NCSC states plainly that these selectors are liable to change, so behavioural detection of DNS hijacking outlives the indicator list.

7
Re-run remediation against CISA KEV, including hardware you do not own. CVE-2023-50224 was KEV-listed on September 3, 2025 with a September 24, 2025 due date and was still being exploited in 2026.

Extend KEV-driven remediation reviews to cover the categories of equipment your workforce uses to reach corporate systems, not only the assets in your CMDB.

References

SOURCES

  • NCSC, "APT28 exploit routers to enable DNS hijacking operations," Version 1, April 7, 2026 - https://www.ncsc.gov.uk/news/apt28-exploit-routers-to-enable-dns-hijacking-operations and https://www.ncsc.gov.uk/sites/default/files/2026-04/NCSC_APT28-exploit-routers-to-enable-DNS-hijacking-operations.pdf
  • NCSC press release, "UK exposes Russian military intelligence hijacking vulnerable routers for cyber attacks," April 7, 2026 - https://www.ncsc.gov.uk/news/uk-exposes-russian-military-intelligence-hijacking-vulnerable-routers-for-cyber-attacks
  • FBI IC3 Public Service Announcement I-040726-PSA, "Russian GRU Exploiting Vulnerable Routers to Steal Sensitive Information," April 7, 2026 - https://www.ic3.gov/PSA/2026/PSA260407
  • Microsoft Security Blog, "SOHO router compromise leads to DNS hijacking and adversary-in-the-middle attacks," April 7, 2026 - https://www.microsoft.com/en-us/security/blog/2026/04/07/soho-router-compromise-leads-to-dns-hijacking-and-adversary-in-the-middle-attacks/
  • National Vulnerability Database, CVE-2023-50224 - https://nvd.nist.gov/vuln/detail/CVE-2023-50224
  • Zero Day Initiative advisory ZDI-23-1808 - https://www.zerodayinitiative.com/advisories/ZDI-23-1808/
  • EUR-Lex, Regulation (EU) 2016/679, Article 83(4) and 83(5) - https://eur-lex.europa.eu/eli/reg/2016/679/oj
  • Data Protection Act 2018, section 157 - https://www.legislation.gov.uk/ukpga/2018/12/section/157
  • CISA Known Exploited Vulnerabilities Catalog, CVE-2023-50224 added September 3, 2025 - https://www.cisa.gov/known-exploited-vulnerabilities-catalog
  • The Register, "Russia's Fancy Bear still attacking routers to boost fake sites, NCSC warns," by Connor Jones, April 7, 2026 - https://www.theregister.com/2026/04/07/russia_fancy_bear_ncsc_router_attack/
  • The Hacker News, "Russian State-Linked APT28 Exploits SOHO Routers in Global DNS Hijacking Campaign," April 7, 2026 - https://thehackernews.com/2026/04/russian-state-linked-apt28-exploits.html