Skip to content
NewCybercrime Report 2026 · Eight Chapters · A Post Is Not An Incident → Explore The ChaptersCybercrime Report 2026 → Explore
All full reports
Full reportDelivery infrastructureMonograph 12 / 1918 min read

Before the Payload Lands: Dual-Use Gates, RMM Servers and Relays

How Keitaro traffic gates, TacticalRMM, Hak5 Cloud C2, FRP and ReverseSSH can stage access, and how to tell them from legitimate use

By
Odin by Brandefense
Published
19 November 2026
Reading time
18 min read
Edition
Version 1.1 · TLP:CLEAR
Scope
Prepared November 2026

1. Executive Summary

In short

In a two-week observation window ending 18 November 2026, Odin by Brandefense studied five dual-use tools that attackers can stage before a payload lands. Of 5,892 endpoints, an estimated 526 deserve attention. Each tool also serves marketers, IT providers or developers, so most hosts stay on watchlists. Three hosts are confirmed adversary infrastructure. Block those three, allowlist your approved remote-management domains, and alert when a browser visit is followed by an installer launch.

The five tools are Keitaro, a commercial traffic direction system (TDS), and TacticalRMM, an open-source remote monitoring and management (RMM) suite. The other three are Hak5 Cloud C2, a controller for Hak5 implant devices, and two tunneling tools, Fast Reverse Proxy (FRP) and ReverseSSH. None of them is malware. Each has legitimate users, and each can also give an attacker a step in an access chain.

1

An estimated 57 Keitaro gates look like cloaking gates, not marketing.

Only an estimated 57 Keitaro servers fit cloaking-gate patterns; most Keitaro installs serve ordinary marketing. They fall into three clusters: templated CN=SolidityGameServer servers, fresh default-certificate VPS gates, and fake CDN or tracker domains. Their redirects are visible, but not where they lead, so all stay on watchlists.

2

TacticalRMM on fresh disposable domains offers signed, persistent remote access.

An estimated 338 TacticalRMM servers sit on new .online, .xyz, .top or duckdns[.]org domains. A signed MeshCentral agent gives an operator a shell over HTTPS port 443. Hostile use is not shown, so these stay on watchlists.

3

Relays hide backends; three hosts are confirmed adversary infrastructure.

We found 114 FRP relays, and 86 of them (75.4%) expose a dashboard on port 7500. Four open directories held relay configs, a key dump and an exploit playbook. Only three hosts pass the bar for blocking: a Hak5 Cloud C2 server next to GoPhish, and two TLS-wrapped ReverseSSH listeners.

Three Decisions for Leaders

  1. Allowlist remote-management tools by domain. Block unapproved TacticalRMM, MeshCentral and out-of-band SSH management traffic by default. Permit only documented corporate or managed service provider (MSP) subdomains.
  2. Link browser clicks to installer launches. Join workstation connections to unapproved redirector or disposable domains with Windows Installer, PowerShell or RMM agent launches within five minutes.
  3. Never block a dual-use banner on its own. Require a second signal, such as a co-hosted phishing kit or an IP-literal reverse-shell certificate. Hold everything else on watchlists, and ignore matches on mail ports.

The Study in Numbers

NumberWhat it is
5,892Deduplicated endpoints in the study set (an estimate from aggregate summaries)
526Estimated priority candidates and relays across the five tools
57Keitaro cloaking-gate candidates
338TacticalRMM fleet candidates on disposable or self-signed setups
114FRP relays; 86 (75.4%) expose the port 7500 dashboard
3Hosts confirmed as adversary infrastructure and published for blocking

How to use this report.

Security leaders can read Chapter 1, Chapter 6 and Chapter 7. SOC and detection engineers should start with Chapters 7 to 9. CTI analysts and infrastructure teams will find the tool detail in Chapters 2 to 6.

2. How the Pieces Can Fit

In short

We measured the five tool populations separately. The four-layer access chain below is an analytical model of how they can combine, not one observed campaign.

An analytical access model

Many initial-access operations split their work across separate servers. One server attracts traffic, another filters it, a third keeps access to the victim, and a fourth relays traffic out of sight. The five tools in this report map onto those roles.

Figure 1: Four functional tool layers can decouple edge traffic filtering from backend relays
Figure 1: Four functional tool layers can decouple edge traffic filtering from backend relays
LayerRoleToolsWhat it gives an operator
0 LuresBring browsers inSearch ads, SEO lures, phishing linksRaw HTTP clicks
1 Traffic gateFilter visitors and redirect the chosen onesKeitaro TDS, Click APIHTTP 302 redirects that skip sandboxes
2 Remote managementKeep access to an enrolled hostTacticalRMM, MeshCentral, wildcard TLSA signed agent with a WebSocket shell on 443
3 RelaysHide the backend behind NAT-crossing tunnelsfrps on 7000, frpc.toml, Hak5 on 2022, ReverseSSHSOCKS5 and SSH tunnels across cheap VPS hops

This is a model. Outside a few shared servers, such as Hak5 Cloud C2 and GoPhish on 146.190.211[.]163, nothing links a specific Keitaro gate to a specific TacticalRMM or FRP server.

Populations and candidates

These tools are popular with legitimate users, so most installs are benign. The table gives each population and its estimated candidates.

ToolLayerEndpoints (hosts)Estimated candidatesMain ports and hosting
Keitaro TDSTraffic gate3,092 (2,368)57443, 80; DigitalOcean and Vultr (US, NL, DE)
TacticalRMMRemote management2,634 (2,157)338443, 1024, 4430; Hetzner and OVH (US, DE, BR)
FRP (frps)Relay114 (114)1147500, 7000; Alibaba and Tencent (CN 70.2%)
Hak5 Cloud C2Relay and implant control22 (16)118080, 9090 plus 2022; AWS and DigitalOcean (US, NL)
ReverseSSHRelay30 (23)63232, 8082; 8 POP3S 995 mail collisions removed

The 5,892 endpoint and 526 candidate totals are estimates from aggregate summaries. Endpoint-level records for these populations are withheld because most hosts are dual-use. Chapter 9 lists a selected set of 28; only its 3 confirmed endpoints are published, and 25 are withheld.

"Software confirmed" names the tool. It does not mean the operator is hostile.

3. Traffic Gates: Keitaro TDS

In short

Keitaro decides in milliseconds which visitor sees which page. An estimated 57 of 3,092 Keitaro endpoints look like cloaking gates, but none shows where its redirect leads, so all stay on watchlists.

How a cloaking gate works

Keitaro was built for affiliate marketing. It reads each visitor's network (ASN), country, operating system and browser, then routes the visitor to one of several weighted campaign streams. An attacker can use the same filters to send sandboxes and security scanners to a harmless page, while real targets get a redirect to an installer.

Figure 2: Signed RMM agents and reverse TCP multiplexers replace custom malware loaders
Figure 2: Signed RMM agents and reverse TCP multiplexers replace custom malware loaders

Operators run Keitaro in two ways:

  • Direct redirectors answer with HTTP 302 and a Set-Cookie: keitaro header. They listen on 443 (1,802 endpoints) and 80 (1,218 endpoints).
  • Backend Click API v3 engines sit behind WordPress or malvertising landing pages. Those pages call /click_api/v3 and receive an X-KClick-Token header.

The gate sits upstream of the payload. A defender who sees the 302 chain can act before an installer lands.

Three clusters that stand apart from marketing

Commercial agencies usually bind Keitaro to aged brand domains behind Cloudflare (219 such records). The 57 candidates look different. They fall into three clusters.

ClusterCertificates seenEstimated gatesHosting and fingerprint
Default CN=0.0.0.0 TLS97321Newly issued self-signed certificate, HTTP 302 with X-KClick-Token, JARM prefix 2ad2ad16d2ad2ad00042
SolidityGameServer dual-port4524Paired 8081 and 8082 listeners, mostly on The Constant Company in Germany, JARM prefix 2ad2ad16d00000022c2a
CDN and tracker masqueradeShort-lived12Privacy or offshore hosts such as Privex Inc. and Melbikomas UAB, on domains that mimic ad or CDN services

The SolidityGameServer gates share certificate names, JARM values and one provider. The sampled ones were first seen on 13 and 14 November 2026. That uniformity suggests templated or automated setup rather than manual work.

Why these gates are not blocked

A gate's visible response shows the 302 and the Click API markers. It does not show the downstream .msi or .ps1 destination. Without that, a gate cannot be told apart from an aggressive ad tracker. All eight sampled Keitaro gates in Chapter 9 are on watchlists, and their addresses are withheld.

4. Persistence: TacticalRMM and Hak5 Cloud C2

In short

Signed remote-management agents get past controls that would stop an unsigned loader. An estimated 338 TacticalRMM servers and 11 Hak5 Cloud C2 controllers fit the pattern, and one Hak5 server shares its host with a phishing kit.

Living off the RMM

Endpoint protection blocks many unsigned commodity loaders. A signed, legitimate RMM installer pointed at an attacker's own server raises fewer alarms. CISA advisory AA23-025A describes this abuse of remote monitoring and management software.

Figure 3: Synthesized protocol model maps HTTP 302, RMM WebSocket, and FRP handshakes to detections
Figure 3: Synthesized protocol model maps HTTP 302, RMM WebSocket, and FRP handshakes to detections

Figure 3 is a protocol model built from the public specifications. It shows where each step appears on the wire and which rule sees it. No single host was seen doing all of these steps.

How a TacticalRMM server is laid out

A standard TacticalRMM install uses three DNS names behind Nginx. The official install covers them with one Let's Encrypt wildcard certificate (*.<domain>), though custom certificates also work.

HostnameServiceCommon portsRecords
rmm.<domain>Vue operator console443Not split out
api.<domain>Django uWSGI REST backend; agents enroll through POST /agents/install/443 or 1024405
mesh.<domain>MeshCentral relay; remote shells and desktops run through WSS /meshrelay.ashx443 or 4430232

Once enrolled, the signed MeshAgent.exe opens a WebSocket over port 443. That gives an operator PowerShell and desktop access that looks like normal HTTPS.

Disposable domains versus MSP domains

Legitimate MSPs run TacticalRMM on corporate domains that are years old. The 338 candidates instead sit on newly registered disposable domains and dynamic DNS services.

Eight sampled servers sit on fresh wildcard certificates. Five are shown below, with addresses and domains withheld.

Certificate patternHosting provider (country)What stands out
*.<withheld>.onlineRackvolt Cloud (US)New offensive-themed wildcard certificate; API on 1024 as well as 443
*.<withheld>.topM247 Europe SRL (US)New .top wildcard on a commercial VPS network
*.<withheld>.xyzSollutium EU (NL)New .xyz wildcard on a commercial VPS network
*.<withheld>.duckdns.orgBlatant Tech (SC)New dynamic DNS wildcard on an offshore VPS
*.<withheld>.onlineCherry Servers (US)New wildcard exposing both the 1024 API and the 4430 relay

The certificates are single wildcards. We did not see an explicit three-name certificate list, and we saw no interaction with a victim. All eight stay on watchlists.

Hak5 Cloud C2 and a phishing neighbor

Hak5 Cloud C2 manages Hak5 hardware implants, such as the LAN Turtle, Packet Squirrel and WiFi Pineapple. It also manages Linux persistence clients (cc-client) set up through /etc/device.config. A controller shows a web console on 8080, 9090, 8088 or 443, plus port 2022 for device SSH tunnels.

Of 22 Hak5 endpoints on 16 IPs, an estimated 11 are controllers. Port 2022 belongs to Hak5 itself, so a console plus 2022 alone is not enough to block. The two standalone controllers in Chapter 9, on Amazon AWS and DigitalOcean, stay on watchlists.

One server is different. 146.190.211[.]163 on DigitalOcean (US) runs Hak5 Cloud C2 on 8088 and 2022. The same host runs a separate GoPhish phishing server on 3333, with the certificate CN=*.maquitacushunfin[.]site. A phishing kit next to an implant controller is a second, independent offensive signal. This host is confirmed and published for blocking.

5. Relays: FRP and ReverseSSH

In short

Reverse proxies let an operator reach inside a network through a cheap cloud relay instead of a core server. We found 114 FRP relays, four open directories with relay files, and two TLS-wrapped ReverseSSH listeners confirmed for blocking.

How FRP shields the backend

FRP has two parts. The server, frps, runs on a cloud VPS. The client, frpc, runs inside the target network and dials out. Traffic then flows back through the relay, so the operator's own machine never faces the target. FRP can carry RDP, SMB and SOCKS5 traffic.

Figure 4: Three Let's Encrypt subdomains and an FRP control port expose the relay stack
Figure 4: Three Let's Encrypt subdomains and an FRP control port expose the relay stack
PortServiceWhat a defender sees
7000frps control port; streams multiplexed with yamuxA JSON login with version, privilege_key, timestamp and run_id, but only when TLS is off
7500 (sometimes 17500)frps web dashboard, with the /api/proxy/tcp APIAn HTTP dashboard listing active tunnels

The privilege_key is an MD5 digest of the auth.token and a timestamp. From FRP v0.50.0, transport TLS is on by default (transport.tls.enable = true). The plaintext login is visible only on older builds, when TLS is turned off, or behind TLS inspection.

Where the relays run

Of the 114 relays, 86 (75.4%) leave the 7500 dashboard reachable. China hosts 70.2% of them, led by 61 nodes on Alibaba Cloud and 9 on Tencent. Developers also use FRP to expose test services, and ports and providers alone cannot tell the two apart. All 114 relay addresses are withheld and held on watchlists.

What the open directories held

Four open HTTP directories exposed five relay-related files. Some sit next to a 7000 relay. Whether their owners are attackers or lab administrators is unknown, and no separate C2 framework was found on them. All four hosts stay on watchlists.

Exposed filePortHosting (country)What it is
/frpc_kd2.toml8000Huawei Cloud (CN)FRP v0.5x client config with an auth.token (secret withheld)
/poll_frpc.sh18080Alibaba Cloud (CN)Cron watchdog script that restarts frpc tunnels when they drop
/keydump.txt8000Oracle Cloud (SG)Dump of extracted keys, next to a 7000 relay
/blind-rce-exploit/PLAYBOOK.md9999Tencent (CN)Remote code execution playbook next to a 7000 relay (2 records)

Three of these records were first listed before our window opened. The frpc_kd2.toml record was first seen on 17 November 2026.

ReverseSSH on non-standard ports

ReverseSSH is a static binary that accepts reverse SSH shells and port forwards over firewall-friendly ports. We saw 30 records on 23 IPs. Of these, 11 listen on the default port 3232: 4 as plaintext SSH and 7 behind self-signed CN=:3232 TLS certificates.

Plaintext listeners on 3232, 8082 or 8443 send their SSH banner (SSH-2.0-Go or SSH-2.0-OpenSSH) straight to the client. Suricata SID 9512008 catches that.

The two hosts confirmed for blocking hide the banner inside TLS. Each uses a certificate whose common name is its own IP:port:

EndpointHosting (country)CertificateJARM prefix
77.111.101[.]84:8082 (also 8443)Latitude.sh (BR)CN=77.111.101[.]84:80823fd3fd0003fd3fd00043
165.154.199[.]52:443Scloud Pte Ltd (SG)CN=165.154.199[.]52:4433fd3fd0003fd3fd00043

TLS hides the inner SSH handshake from wire signatures. Find these hosts through the IP-literal certificate and JARM, egress hunting in KQL, and ssh -R process events in Sigma.

6. Telling Hostile From Legitimate

In short

Blocking every Keitaro or TacticalRMM server would break marketing analytics and outsourced IT support. Four common lookalikes explain why most hosts in this report stay on watchlists.

An exposed TacticalRMM login page shows that the software is present. It does not show intent.

Four lookalikes

LookalikeWhat we sawHow it is handled
MSPs and marketers158 TacticalRMM records on aged MSP domains; Keitaro behind enterprise CDNsSet aside and logged only, when the domain is old, the certificate is clean and no offensive tool shares the host
Mail servers8 ReverseSSH records were POP3S mail servers at an Italian hosting provider (ports 995 and 110); 5 Hak5 records sat on mail ports of a Dutch VPSMatches on mail ports 110, 143, 465, 993 and 995 are ignored
Authorized testersTwo suspected Hak5 Cloud C2 instances on Microsoft Azure (IE) and Amazon AWS (DE) present CN=pentest certificatesAnalyst review; never auto-blocked or published
Home broadbandOne TacticalRMM host sits on a dynamic Brazilian subscriber network; 30 TacticalRMM records are on dynamic ISPsWatchlist entries that expire after 24 hours, so reassigned customer addresses are never blocked for good

What it takes to block

Everything else in this report needs a second, independent signal before a block. The three confirmed hosts have one each:

  • A co-hosted offensive tool. 146.190.211[.]163 pairs Hak5 Cloud C2 with GoPhish.
  • An IP-literal reverse-shell certificate. 77.111.101[.]84 and 165.154.199[.]52 wrap SSH in TLS under their own IP:port names.

A captured installer redirect from a Keitaro gate would also qualify. None of the sampled gates showed one.

7. What Defenders Should Do

Within 24 hours

  • Block the three confirmed hosts: 146.190.211[.]163:8088, 77.111.101[.]84:8082 and 165.154.199[.]52:443.
  • Keep watchlist indicators (Keitaro gates, wildcard TacticalRMM servers, standalone Hak5 controllers, FRP open directories, CN=pentest hosts) out of automatic block feeds.
  • Exclude mail-port matches (995 POP3S and the other mail ports) from ReverseSSH hunting.

Within 72 hours

  • List every approved RMM tool. Allow rmm.*, api.* and mesh.* traffic, including /meshrelay.ashx, only to documented corporate or MSP names, at DNS and the TLS proxy.
  • Add WDAC or AppLocker rules that block unapproved tacticalrmm.exe, MeshAgent.exe and frpc.exe, even when the files carry a valid Authenticode signature.
  • Restrict workstation egress on relay ports 2022, 3232, 4430, 7000, 7500 and 17500.

Within 7 to 30 days

  • Inspect HTTP 302 chains with keitaro cookies. Correlate browser visits to domains registered in the last 30 days on .xyz, .top, .online or duckdns[.]org with .msi, .vbs or .ps1 launches.
  • Deploy Suricata SIDs 9512001 to 9512008, the Sentinel KQL hunt and the three Sigma rules.
  • Alert on plaintext SSH-2.0 banners from relay ports and on IP-literal TLS certificates.
  • Rotate any token found in an exposed frpc.toml.

8. Detection Rules

The full rules are in the Detection rules tab. Each file targets one layer of the model.

Rule fileWhat it catches
odin_keitaro_tacticalrmm_frp.rulesSuricata 7: eight signatures for Keitaro redirects, TacticalRMM enrollment and relay traffic, plaintext FRP logins, Hak5 consoles and ReverseSSH banners
sentinel_unapproved_rmm_and_tds_redirect_chain.kqlMicrosoft Sentinel and Defender for Endpoint: browser-to-installer chains, unapproved RMM or FRP processes, and relay egress
sigma_tacticalrmm_frpc_unauthorized_execution.ymlSigma (3 rules): unapproved TacticalRMM or MeshCentral agents, frpc tunnels, and Hak5 or ReverseSSH tunnels
odin_tds_rmm_frp_staging_configs.yarYARA (4 rules): Keitaro gate configs, TacticalRMM bootstrap manifests, FRP client configs, and Hak5 or ReverseSSH provisioning files

How the Suricata signatures work

SIDLooks forTactic
9512001A server 302 response whose cookie contains keitaro and a subid, _subid or stream markerInitial Access
9512002An internal client requesting /click_api/v3Initial Access
9512003An internal client calling TacticalRMM enrollment at /agents/install/Persistence
9512004A WebSocket upgrade to /meshrelay.ashxCommand and Control
9512005An outbound plaintext FRP login with privilege_key, version and timestamp or run_idCommand and Control
9512006Access to the FRP dashboard API /api/proxy/tcp on 7500, 17500, 8080 or 9527Command and Control
9512007Hak5 Cloud C2 branding in an HTTP response body (unmodified consoles only)Command and Control
9512008A server response starting SSH-2.0-Go or SSH-2.0-OpenSSH from ports 3232, 8082 or 8443Command and Control

The HTTPS rules (9512001 to 9512004 and 9512007) need TLS inspection or offload. The TCP rules (9512005 and 9512008) belong on perimeter egress. SID 9512005 sees only FRP builds older than v0.50.0 or sessions with TLS turned off. Bypass 9512003 and 9512004 for your documented MSP api. and mesh. names, and exclude sanctioned jump hosts from 9512008.

How the KQL query works

The Sentinel query uses a 14-day lookback and has three branches.

  1. Browser to installer. It joins a browser connection to a confirmed host, a click_api/v3 URL or a disposable domain with a later process launch. It looks for msiexec.exe, PowerShell, script hosts, tacticalrmm.exe or MeshAgent.exe on the same device within five minutes.
  2. Unapproved RMM or FRP process. It flags tacticalrmm.exe, MeshAgent.exe or frpc, or command lines that mention meshrelay.ashx, /agents/install/, frpc.toml, frpc.ini, privilege_key or device.config.
  3. Relay egress. It flags connections to the three confirmed hosts or to rmm., api. or mesh. names on disposable domains. It also flags RMM, FRP or SSH processes that reach relay ports 2022, 3232, 4430, 7000, 7500 or 17500.

Fill the ApprovedMspDomains list before use. Treat the disposable-domain branch as a hunting lead, not an isolation trigger. Endpoint network events do not carry certificate details, so check domain age and certificate names in certificate transparency or passive TLS data first.

How the Sigma and YARA rules work

RuleFires when
Sigma: Unapproved TacticalRMM or MeshCentral Agent Executiontacticalrmm.exe or MeshAgent.exe runs, or a command line names RMM enrollment or relay paths, outside approved MSP names
Sigma: FRP Reverse Relay Client Execution or Tunnel Multiplexingfrpc runs, or a command line references frpc.toml or serverPort = 7000; this works even when FRP traffic is encrypted
Sigma: Hak5 Cloud C2 or ReverseSSH Outbound Tunnel InvocationA Linux command references Hak5 /etc/device.config, or ssh -R targets port 2022, 3232, 8082, 8443 or 443
YARA: ODIN_Keitaro_TDS_Cloaking_Gate_Configkeitaro appears with X-KClick-Token, click_api/v3 or object=campaigns, plus stream_id= or 302 Found
YARA: ODIN_TacticalRMM_Agent_Bootstrap_ManifestTwo TacticalRMM markers (/agents/install/, /meshrelay.ashx, tacticalagent) appear with two of the api., mesh. and rmm. prefixes
YARA: ODIN_FRP_Reverse_Relay_Client_ConfigA file has a server address, serverPort = 7000, an auth.token or privilege_key, and a socks5 or [[proxies]] entry
YARA: ODIN_Hak5_CloudC2_And_ReverseSSH_ProvisioningTwo Hak5 markers appear with 2022, or device.config appears with :3232

How to sort a host

The triage logic turns these signals into an action. It never blocks on one dual-use signal.

What the host showsAction
An approved MSP or enterprise assetAllow
A match on a standard mail portIgnore as a collision
An aged domain, a clean business certificate and no co-hosted offensive toolLog only
A CN=pentest certificateAnalyst review
A dynamic subscriber ISP addressWatchlist for 24 hours
A Keitaro gate with no captured installer, a wildcard RMM server on a new disposable domain, a standalone Hak5 controller, or an FRP relay or open directoryWatchlist and analyst review
A co-hosted offensive tool, a captured installer redirect, or an IP-literal reverse-shell certificateBlock

Before you deploy

Only the KQL query carries an allowlist, and it matches exact hostnames. The Sigma, Suricata and YARA rules have none, so suppress sanctioned MSP agents by exact host and agent ID first. Every rule file loads without errors in its engine (Suricata 7.0.12, YARA-X, pySigma, Kusto.Language); the rules were not replayed against live traffic. The YARA rules matched four sample config files and did not fire on two benign controls. Test all rules in your own environment before production use.

9. Indicators

The Indicators tab holds the 3 confirmed endpoints. The other 25 records in the selected set are withheld.

TypeCount
Confirmed adversary endpoints to block (IP and port)3
Watchlist endpoints, addresses withheld24
Excluded mail-server collision, address withheld1

Confirmed endpoints to block

EndpointToolHosting (country)Key fingerprintFirst seen
146.190.211[.]163:8088Hak5 Cloud C2DigitalOcean (US)Hak5 console on 8088 and 2022; GoPhish on 3333 with CN=*.maquitacushunfin[.]site13 Nov 2026
77.111.101[.]84:8082ReverseSSHLatitude.sh (BR)TLS-wrapped SSH on 8082 and 8443, CN=77.111.101[.]84:8082, JARM prefix 3fd3fd0003fd3fd0004314 Nov 2026
165.154.199[.]52:443ReverseSSHScloud Pte Ltd (SG)TLS-wrapped SSH, CN=165.154.199[.]52:443, same JARM prefix13 Nov 2026

Watchlist entries

These 24 entries are for enrichment and analyst triage, not for blocking. Addresses and domains are withheld.

GroupEntriesHosting (country)Fingerprint to hunt
Keitaro gates8The Constant Company (DE, 3), Ultahost (DE), ROUTE 95 LLC (NL), Privex Inc. (NL), Melbikomas UAB (DE), Mulgin A.S. (UA)CN=SolidityGameServer on 8081, 8082 and 443; CN=0.0.0.0 with X-KClick-Token; .life, .site and lookalike domains with /click_api/v3
TacticalRMM servers on new wildcards8Rackvolt Cloud (US), M247 Europe (US), Sollutium EU (NL, 2), Blatant Tech (SC), Ultahost (NL), Amazon AWS (US), Cherry Servers (US)New wildcards on .online, .top, .xyz, .art, .site and duckdns[.]org; API on 1024; one with /meshrelay.ashx on 443
TacticalRMM on dynamic broadband1Dynamic telecom network (BR)24-hour watchlist only
Standalone Hak5 controllers2Amazon AWS (US), DigitalOcean (US)Console on 8080 or 9090 plus 2022
FRP open directories4Tencent (CN), Oracle Cloud (SG), Alibaba Cloud (CN), Huawei Cloud (CN)The relay files listed in Chapter 5
CN=pentest host1Microsoft Azure (IE)CN=pentest with 3389 open; the second CN=pentest host (AWS, DE) is outside the selected set

Most entries were first seen between 13 and 18 November 2026. Three FRP open-directory records were first listed before our window.

Withheld: all 114 FRP relay addresses, the 24 watchlist addresses and domains, the frpc_kd2.toml token, the mail-server collision host, and all other candidate addresses.

ATT&CK tactics

TacticHow the tools fitDetection
Initial AccessKeitaro gates filter visitors and redirect the chosen onesSIDs 9512001 and 9512002; KQL branch 1; YARA Keitaro rule
PersistenceSigned TacticalRMM and MeshCentral agents enroll and hold accessSID 9512003; KQL branch 2; Sigma rule 1; YARA TacticalRMM rule
Command and ControlMeshCentral WebSockets, FRP tunnels, Hak5 SSH and ReverseSSH listenersSIDs 9512004 to 9512008; KQL branch 3; Sigma rules 2 and 3; YARA FRP and Hak5 rules

10. Scope and Limits

  • Window. The findings cover a two-week observation window ending 18 November 2026. Three FRP directory records were first listed before the window; no dates are given for them.
  • Population totals are estimates. The 5,892 endpoints, 526 candidates, 57 gates and 338 TacticalRMM servers come from aggregate summaries. Only 3 confirmed endpoints are published at endpoint level.
  • The access chain is a model. The five tools were measured separately. Apart from shared hosts such as the Hak5 and GoPhish server, nothing links one layer to another.
  • Software present is not hostile intent. Most hosts run legitimate tools for marketing, IT support or development. Only three hosts are confirmed adversary infrastructure.
  • Redirect targets are unseen. No sampled Keitaro gate showed where its redirect leads, and no installer download was seen.
  • Open directories show staging, not use. The relay configs, key dump and playbook are staged files. Whether their owners are attackers or lab administrators is unknown.
  • Rule testing. Every published rule file loads without errors in its engine (Suricata 7.0.12, YARA-X, pySigma and Microsoft's Kusto.Language analyzer, checked on 20 November 2026). The rules were not replayed against live traffic, so test and tune them in your own environment before production use.

11. Sources

  1. MITRE ATT&CK Enterprise tactics: Initial Access, Persistence, Command and Control. https://attack.mitre.org/tactics/enterprise/ (verified 18 November 2026)
  2. CISA. Joint Cybersecurity Advisory AA23-025A, Protecting Against Malicious Use of Remote Monitoring and Management Software. https://www.cisa.gov/news-events/cybersecurity-advisories/aa23-025a (verified 18 November 2026)
  3. Keitaro TDS documentation: campaign streams, filters, redirects and Click API v3. https://docs.keitaro.io/ (verified 18 November 2026)
  4. TacticalRMM documentation, How It All Works: rmm, api and mesh hostnames, wildcard TLS, Django and MeshCentral. https://docs.tacticalrmm.com/howitallworks/ (verified 18 November 2026)
  5. fatedier. Fast Reverse Proxy (FRP) repository: frps, frpc.toml, default TLS from v0.50.0, token authentication. https://github.com/fatedier/frp (verified 18 November 2026)
  6. Hak5 Cloud C2 documentation: deployment, ports and device.config. https://docs.hak5.org/cloud-c2/ (verified 18 November 2026)

What is…?

Short explanations of the tools and families named in this study.

All explanations

Frequently asked questions

What are these tools, and why do attackers want them?

Each tool fills one step of an access chain. We measured the five populations separately, so the chain below is a model, not one observed campaign. The appeal is simple. Security tools often stop unsigned malware. They are less likely to stop a signed remote-management agent or a marketing redirect.

How many look hostile?

Very few, once legitimate use is set aside. The 57 Keitaro candidates fall into three clusters. Some use a templated certificate named SolidityGameServer on paired ports. Some use fresh default certificates on cheap VPS hosts. Others sit on domains that imitate ad or CDN services. Their redirects are visible, but not where they lead.

Which hosts should be blocked?

Only three hosts carry a second, independent sign of hostile use: Everything else belongs on a watchlist. A Keitaro gate may be an aggressive marketer. A TacticalRMM server may be a small IT provider. An FRP relay may be a developer's test tunnel. Some lookalikes are plainly benign: eight "ReverseSSH" matches were mail servers, and two Hak5 servers carry CN=pentest certificates that suggest authorized testing.

How do defenders catch this without breaking IT?

Start from what your organization approves. If only your MSP's RMM domains are allowed, any other TacticalRMM or MeshCentral traffic stands out at once. Then link events across the chain. A browser visit to a fresh disposable domain, followed within five minutes by msiexec.exe, PowerShell or an RMM agent, is a strong lead. Our Sentinel query does this join.

Defender kit

Indicators of compromise

Verified adversary infrastructure and tool hashes from this study. Every value below is also discussed in the report.

IDTypeIndicatorPortContextAction
R12-IOC-01 IPv4 146.190.211[.]163 8088 Hak5 Cloud C2 controller Block
R12-IOC-02 IPv4 77.111.101[.]84 8082 ReverseSSH listener Block
R12-IOC-03 IPv4 165.154.199[.]52 443 ReverseSSH listener Block

25 records from this study are not published in this table: candidate, suspected and dual-use hosts, victim-related values, and hashes of exposed secrets or files.

Defender kit

Detection rules

The rule files exactly as printed in the report. Test them in your own engine before deployment.

Suricata odin_keitaro_tacticalrmm_frp.rules 29 lines
Download
YARA odin_tds_rmm_frp_staging_configs.yar 80 lines
Download
Sentinel KQL sentinel_unapproved_rmm_and_tds_redirect_chain.kql 73 lines
Download
Sigma sigma_tacticalrmm_frpc_unauthorized_execution.yml 89 lines
Download