1. Executive Summary
In short
In a two-week observation window ending 18 November 2026, Odin by Brandefense mapped 2,674 endpoints on 2,468 hosts that carry the ARTEX or DarkSword family labels. Of these, 526 endpoints on 516 hosts are confirmed command-and-control (C2) servers. ARTEX controllers reuse one certificate per hosting provider. DarkSword gathers on CNServer LLC. Nothing confirmed links the two families. Block the confirmed endpoints in the Indicators tab and hunt non-browser traffic to ports 8787, 8790, 8888 and 2000.
ARTEX and DarkSword are two high-volume C2 family labels. Both sit on cheap virtual private servers (VPS) leased in bulk from a few regional resellers. This report maps where they sit, what they reuse, and what is safe to block.
Two providers host 82.7% of confirmed ARTEX controllers.
YANCY LIMITED holds 237 confirmed ARTEX controllers and PEG TECH INC holds 175. Every controller in each group presents the same TLS certificate and almost always the same JARM fingerprint. Around them sit 1,272 unconfirmed listeners on port 8787 and 108 on port 8790.
DarkSword puts over half its endpoints on one provider.
CNServer LLC hosts 189 of the 362 DarkSword endpoints (52.2%). Of these, 186 are suspected listeners on port 443, and 174 share one JARM value. The 28 confirmed DarkSword endpoints sit elsewhere, on 19 hosts that often pair ports 80, 443, 8443, 8888 and 2000.
Most endpoints should be watched, not blocked.
Only 526 of the 2,674 endpoints (19.7%) are confirmed. The rest include shared Cloudflare and Traefik certificates, open directories, dynamic telecom addresses and uncorroborated listeners. Blocking them outright would hit shared infrastructure and unrelated tenants.
Three Decisions for Leaders
- Block confirmed controllers and review the list every 30 days. Start with the Indicators tab. Leased VPS addresses change hands after an operator rotates, so a block that never expires will age badly.
- Hunt non-browser egress to the cluster ports. Ports 8787, 8790, 8888 and 2000 carry most of the fleet. Alert when a script, shell or service process talks to them across these hosting networks.
- Ask for a second signal before blocking shared infrastructure. Shared CDN certificates, mail ports and telecom subscriber ranges belong on short watchlists until a URI, certificate or process match backs them up.
The Study in Numbers
| Number | What it is |
|---|---|
| 2,674 | Endpoints with an ARTEX or DarkSword label, on 2,468 distinct IPv4 hosts |
| 526 | Confirmed C2 endpoints on 516 hosts (19.7%), ready to block |
| 2,312 | ARTEX endpoints on 2,157 hosts; 1,272 of them on port 8787 |
| 362 | DarkSword endpoints on 313 hosts; 189 (52.2%) on CNServer LLC |
| 82.7% | Share of the 498 confirmed ARTEX controllers on YANCY LIMITED and PEG TECH INC (412) |
| 186 | Hosts that expose two to four of these listeners at once |
| 2 | Hosts that carry both families; both are unconfirmed candidates |
How to use this report.
Security leaders can read Chapter 1 and Chapter 7. SOC and detection engineers should start with Chapters 7 to 9. CTI analysts will find the cluster, port and TLS detail in Chapters 2 to 6.
2. Two Families, Four Clusters
In short
The 2,674 endpoints fall into four clusters by family, port and hosting provider. The clusters are an analytical model: shared certificates link hosts inside a cluster, but almost nothing links one cluster to another.
What the labels mean
ARTEX is the family label for ARTEX, the autonomous red-team console described in Monographs 04 and 11. Its default console port is 8787, and it also listens on 443, 80 and 7878. The confirmed YANCY LIMITED and PEG TECH INC controllers in this report are the same certificate clusters counted in those monographs (237 and 175).
"DarkSword" is a family label that comes from an upstream threat feed. We did not tie it to any published vendor report, so the label is not attribution. We use it only to group the endpoints the feed tags that way.
Each endpoint below is an IP, port and family combination.
Counts by family
| Family | Endpoints | Hosts | Endpoints per host | Confirmed | Suspected | Candidate | Other watchlist |
|---|---|---|---|---|---|---|---|
| ARTEX | 2,312 | 2,157 | 1.07 | 498 (497 hosts) | 14 | 1,787 | 12 open directory, 1 dynamic ISP |
| DarkSword | 362 | 313 | 1.16 | 28 (19 hosts) | 212 | 121 | 1 dynamic ISP |
| Total | 2,674 | 2,468 | 1.08 | 526 (516 hosts) | 226 | 1,908 | 14 |
ARTEX makes up 86.5% of the endpoints. The per-family host counts add up to 2,470, two more than the 2,468 distinct hosts. Those two hosts carry listeners of both families (Chapter 5).

The four clusters
| Cluster | What defines it | Size | Confirmed |
|---|---|---|---|
| 1. ARTEX TLS controllers | Confirmed controllers, mostly on ports 443 and 80; 412 sit on YANCY LIMITED and PEG TECH INC behind one shared certificate per provider | 498 confirmed endpoints on 497 hosts | 498 |
| 2. ARTEX port 8787 and 8790 fleet | Listeners on the ARTEX default port and its companion port | 1,380 endpoints (1,272 on 8787, 108 on 8790) | None |
| 3. DarkSword CNServer cluster | Port 443 listeners on CNServer LLC sharing one JARM value | 189 endpoints | None (186 suspected) |
| 4. DarkSword high-port hubs | Listeners on 8888 and 2000, often paired, on CTG Server, cognetcloud and Datacamp | 97 endpoints (69 on 8888, 28 on 2000) | Includes 103.43.71[.]178:2000 |
Inside clusters 1 and 3, one certificate or one JARM value repeats across hundreds of servers. Between clusters, the only shared element is the hosting provider. Two candidate hosts carry both families, and nothing else links ARTEX to DarkSword.
Where the endpoints sit
| Country | Endpoints | Share |
|---|---|---|
| United States | 1,283 | 48.0% |
| Hong Kong | 698 | 26.1% |
| China | 301 | 11.3% |
| Singapore | 89 | 3.3% |
| Japan | 87 | 3.3% |
The Netherlands (34), Germany (31), South Korea (21) and Malaysia (19) follow.
US and Hong Kong locations may reflect where resellers announce their address space rather than where the operators sit. PEG TECH INC, YANCY LIMITED, CNServer LLC, CTG Server Limited and cognetcloud INC all sell VPS capacity in both places.
ATT&CK mapping
| Technique | ID | Where it shows |
|---|---|---|
| Acquire Infrastructure: Virtual Private Server | T1583.003 | Bulk VPS leases on a few regional resellers |
| Non-Standard Port | T1571 | Listeners on 8787, 8790, 8888, 2000, 7676 and 18787 |
| Application Layer Protocol: Web Protocols | T1071.001 | HTTP and HTTPS controllers on 80, 443 and 8443 |
3. ARTEX: Two Certificate Clusters and a Port 8787 Fleet
In short
Confirmed ARTEX controllers listen mostly on port 443 (421) and port 80 (63). The unconfirmed majority sits on port 8787 (1,272) and port 8790 (108), often on the same providers and behind the same certificates.

Ports
| Port | ARTEX endpoints | Note |
|---|---|---|
| 8787 | 1,272 (55.0%) | ARTEX default console port; also RStudio Server's default |
| 443 | 563 | 421 confirmed |
| 80 | 168 | 63 confirmed |
| 8790 | 108 (4.7%) | Companion port, often paired with 8787 |
| 8443 | 19 | |
| 8080 | 16 | |
| 18787 | 14 | Offset variant of 8787 |
| 7878 | 13 | ARTEX default |
| 8899 | 12 | |
| 8081 | 12 | |
| 8788 | 4 |
One certificate per provider
Of the 498 confirmed ARTEX endpoints, 412 (82.7%) sit on two providers. Inside each provider, every controller presents the same TLS certificate. This points to one provisioning template per provider, applied to hundreds of servers.
| Provider | Confirmed controllers | Certificate | JARM | Unconfirmed listeners with the same certificate |
|---|---|---|---|---|
| YANCY LIMITED (US, HK) | 237, all on 443 | One shared certificate on all 237 (100%); certificate name withheld | 29d29d00029d29d00042d43d00041dd469afa8cfbe5e42c631eb3fc55d6787 on all 237 | 29 of 47 port 8787 candidates; 266 endpoints in total carry it |
| PEG TECH INC (US, HK) | 175, all on 443 | One shared certificate on all 175 (100%); certificate name withheld | 3fd3fd16d3fd3fd21c42d42d00000078e356c38fb85055d97a8bed930c8956 on 173 | 167 port 8787 candidates; 343 endpoints in total carry it |
Certificate name withheld for both clusters. The shared certificate ties many unconfirmed port 8787 listeners to confirmed controllers on the same provider.
The other 86 confirmed ARTEX controllers spread across smaller networks. CTG Server Limited holds 9, Hangzhou Alibaba 8, HostPapa 6, NetLab Global 5, Shenzhen Tencent 4 and Oracle 4.
Provider footprint
PEG TECH INC is the largest single home for ARTEX: 634 ARTEX endpoints on 573 hosts (635 endpoints in total). Besides the 175 confirmed controllers, it carries 357 candidates on 8787 and 100 on 8790.
YANCY LIMITED appears under two network names. Together they carry 284 ARTEX endpoints: the 237 confirmed controllers plus 47 port 8787 candidates. They also carry 6 DarkSword endpoints, for 290 in total. In all, 918 ARTEX endpoints sit on PEG TECH INC (634) and the YANCY networks (284).
The port 8787 and 8790 fleet
The 1,380 listeners on 8787 and 8790 are unconfirmed and sit on the same budget providers.
| Provider | Listeners on 8787 or 8790 |
|---|---|
| PEG TECH INC | 357 on 8787, 100 on 8790 |
| Shenzhen Tencent | 102 |
| Hangzhou Alibaba | 77 |
| NetLab Global | 70 |
| cognetcloud INC | 51 |
| Yancy Limited | 47 |
Sixty-two hosts bind 8787 and 8790 together, and 61 of them sit on PEG TECH INC. The pairing looks like a primary and fallback port set by the same template.
Seven port 8787 listeners also serve an open directory, and 41 on 8787 or 8790 sit behind shared Cloudflare or Traefik certificates (Chapter 6).
Confirmed controllers off the two big providers
Several confirmed ARTEX controllers use other ports and providers. Examples include 8.216.54[.]254 on Alibaba US Technology in Japan (ports 80 and 443) and 64.90.1[.]153:8083 on NetLab Global (JARM prefix 1dd40d40d). Another is 110.42.238[.]147:7676 on Shenzhen Tencent. Chapter 9 lists the 12 published ARTEX endpoints in the Indicators tab.
How these counts relate to Monographs 04 and 11
The 237 YANCY and 175 PEG TECH controllers match Monographs 04 and 11. Monograph 04 counted 501 ARTEX endpoints including suspected ones, and Monograph 11 counted 2,121. Each study draws a different set, so the totals differ while the two certificate clusters stay the same.
4. DarkSword: CNServer and High-Port Hubs
In short
DarkSword has 362 endpoints on 313 hosts. Over half sit on CNServer LLC as suspected port 443 listeners. The 28 confirmed endpoints sit on other providers and pair standard ports with 8443, 8888 and 2000.
DarkSword listens on 443 (223 endpoints, 61.6%), 8888 (69, 19.1%), 2000 (28, 7.7%), 80 (23), 3000 (5) and 8443 (3).

Figure 3 is an analytical model inferred from shared certificates, JARM values and port pairs. The HTTP paths in steps 04 and 05 are the strings our rules look for, not captured traffic.
The CNServer LLC cluster
CNServer LLC hosts 189 of the 362 DarkSword endpoints (52.2%). It is the densest single pairing of provider and family in this study.
| CNServer LLC DarkSword group | Endpoints |
|---|---|
| Suspected listeners on 443 | 186 |
...of which share JARM 21d14d00021d21d00042d43d00000091f9827a8676a9d9f27d421962a09b5d | 174 |
| Candidate web consoles on 3000, same JARM | 3 |
A shared JARM value fits either a shared C2 bundle or the provider's standard server image. We therefore keep all 186 port 443 listeners on a 7-day watchlist and do not publish their addresses. Block a subnet here only with a second signal, such as a matching URI.
Confirmed DarkSword hubs
The 28 confirmed DarkSword endpoints sit on 19 hosts, so a confirmed host often exposes more than one port. CTG Server Limited holds 7 of them, Datacamp Limited 4, YANCY LIMITED 3 and SOLLUTIUM 2.
| Endpoint | Provider (country) | Pattern |
|---|---|---|
| 27.124.7[.]56:80, :443, :8443 | CTG Server Limited (HK) | Three confirmed ports on one host; same certificate on 443 and 8443 |
| 103.43.71[.]178:2000 | CyberForest LLC. (KR) | Confirmed listener on port 2000 |
| 193.218.201[.]146:443 | Spartan Host Ltd (US) | TLS controller |
| 206.187.214[.]37:443 | cognetcloud INC (HK) | TLS controller; JARM also seen on Cloudflare-fronted hosts |
| 43.229.112[.]254:443 | YANCY LIMITED (US) | One of 3 confirmed DarkSword endpoints on YANCY |
| 85.137.62[.]160:443 | SOLLUTIUM EU (NL) | One of 2 confirmed in the Netherlands |
| 86.38.247[.]228:80, :443 | Datacamp Limited (SG) | Confirmed on both ports |
Seven confirmed hosts bind ports 80 and 443 together. Six of them are not in the table above. They appear in this report only, not in the Indicators tab.
| Host | Provider (country) |
|---|---|
| 118.107.21[.]193 | CTG Server Limited (HK) |
| 181.215.18[.]87 | Datacamp Limited (HK) |
| 191.40.44[.]243 | CORENET CLOUD SDN. BHD. (US) |
| 47.85.134[.]152 | Alibaba US Technology (US) |
| 52.237.90[.]192 | Microsoft Corporation (SG) |
| 64.90.7[.]155 | NetLab Global (HK) |
Ports 8888 and 2000
DarkSword's 97 high-port listeners split into 69 on 8888 and 28 on 2000. They sit mostly on CTG Server Limited, cognetcloud INC and Datacamp Limited. Nine candidate hosts expose 8888 and 2000 together.
Only one port 2000 listener, 103.43.71[.]178, is confirmed. Treat the rest as leads, because routers and proxies also use these ports (Chapter 6).
5. Multi-Port Hosts and Shared TLS Stacks
In short
Of the 2,468 hosts, 186 (7.5%) expose two to four of these listeners at once. Three JARM values cover 850 endpoints (31.8%), so the TLS stack stays visible even when the addresses rotate.

Port pairs
The 186 multi-port hosts expose 392 endpoints between them. Of these hosts, 170 have two ports, 12 have three and 4 have four.
| Pattern | Hosts | Detail |
|---|---|---|
| ARTEX 8787 + 8790 | 62 | 61 on PEG TECH INC |
| ARTEX 80 + 8787 | 19 | |
| ARTEX 443 + 8787 | 17 | |
| ARTEX 80 + 443 | 12 | Includes confirmed hub 8.216.54[.]254 |
| ARTEX 80 + 443 + 8080 + 8787 | 4 | Four-port hosts |
| DarkSword 80 + 443 | 10 | 7 confirmed two-port hubs plus the three-port hub 27.124.7[.]56 |
| DarkSword 2000 + 8888 | 9 | Candidates |
| DarkSword 443 + 8888 | 7 | |
| DarkSword 443 + 3000 | 3 | All on CNServer LLC |
JARM clusters
JARM is an active TLS fingerprint. Servers that share a TLS library and configuration return the same 62-character value. A shared JARM can mean one operator's template or one provider's default image, so it is a hunting lead, not proof on its own.
| JARM | Endpoints (ARTEX / DarkSword) | Where it concentrates |
|---|---|---|
3fd3fd16d3fd3fd21c42d42d00000078e356c38fb85055d97a8bed930c8956 | 368 (368 / 0) | PEG TECH INC: 173 confirmed 443 controllers plus 195 listeners on 8787 or 80 |
29d29d00029d29d00042d43d00041dd469afa8cfbe5e42c631eb3fc55d6787 | 278 (278 / 0) | YANCY LIMITED: 237 confirmed 443 controllers plus port 8787 candidates |
21d14d00021d21d00042d43d00000091f9827a8676a9d9f27d421962a09b5d | 204 (21 / 183) | CNServer LLC: 174 suspected DarkSword 443 listeners and 3 on 3000; 15 PEG TECH 8787 listeners |
3fd3fd0003fd3fd21c42d42d000000bdfc58c9a46434368cf60aa440385763 | 101 (84 / 17) | Mixed; includes Cloudflare-fronted hosts and confirmed 206.187.214[.]37 |
40d40d40d00040d00042d42d000000831b6af40378e2dd35eeac4e9311926e | 20 (0 / 20) | DarkSword confirmed TLS stack: 11 of the 15 confirmed 443 endpoints, including 27.124.7[.]56 |
The first three values cover 850 endpoints, 31.8% of the study. The fourth mixes in shared Cloudflare edges, so do not block on it.
The two hosts that carry both families
Two hosts carry an ARTEX listener and a DarkSword listener side by side. One is on NetLab Global in Hong Kong (ARTEX on 8787, DarkSword on 8888). The other is on Oracle Corporation in the United States (ARTEX on 8787, DarkSword on 8088).
Both are unconfirmed candidates, so their addresses are withheld. They are the only cross-family link in the data. Two hosts out of 2,468 do not show that one group runs both families.
6. What Not to Block
In short
Port and banner matches on 8787, 443, 8888 and 2000 also hit reverse proxies, data-science servers, routers and telecom subscribers. Of the 2,674 endpoints, 2,148 (80.3%) on 1,952 hosts stay on watchlists for that reason.
Shared CDN and ingress certificates
Forty-three candidate endpoints (41 ARTEX, 2 DarkSword) present a certificate that many unrelated tenants share.
| Certificate subject | Endpoints |
|---|---|
CN=Cloudflare | 26 |
CloudFlare Origin Certificate | 8 |
TRAEFIK DEFAULT CERT | 7 |
Kubernetes Ingress Controller Fake Certificate | 2 |
Blocking one can cut off every site behind the same edge, so route them to a 48-hour review. Do the same for any listener on a mail port (25, 110, 143, 465, 587, 993, 995).
There is one exception. Two ARTEX HTTP controllers on dedicated Alibaba VPS addresses use a Cloudflare Origin Certificate pinned to their own origin. They stay confirmed, because the address belongs to the server, not to a shared edge.
Unrelated services on the same ports
RStudio Server listens on 8787 by default, and Jupyter often uses 8888. MikroTik routers answer on 2000 for bandwidth tests. Enterprise proxies use 8888 as well. A port match alone cannot tell these apart from a C2 listener.
Open directories
Twelve ARTEX candidates also serve a browsable directory index. Seven sit on 8787, three on 8081 inside CNServer LLC (sharing one certificate), one on 8443 and one on 50549. Two of the port 8787 hosts, on POWER LINE DATACENTER in Hong Kong, also expose files that hold secrets. Their addresses are withheld, and they stay on a 7-day watchlist.
Dynamic telecom addresses
Two listeners sit on subscriber telecom networks. One is a confirmed ARTEX controller on China Telecom. The other is a suspected DarkSword listener on JT TELECOM INTERNATIONAL PTE.LTD. in Japan. Addresses on these networks pass between customers, so they get 24-hour watchlists, not permanent blocks.
How the endpoints sort
| Group | Endpoints | Action |
|---|---|---|
| Confirmed on dedicated hosting | 526 (516 hosts) | Block, review every 30 days |
| Suspected | 226 (186 of them DarkSword on CNServer) | Watch for 7 days, look for a second signal |
| Candidates | 1,908 (mostly 8787, 8790, 8888 and 2000 listeners) | Watch for 3 days, look for a URI or process match |
| Open-directory candidates | 12 | Watch for 7 days, hunt with SID 9528004 |
| Dynamic telecom | 2 | Watch for 24 hours |
| Shared CDN or ingress certificate (within the groups above) | 43 | Review for 48 hours; never block automatically |
7. What Defenders Should Do
Within 24 hours
- Block the 20 confirmed endpoints in the Indicators tab and the six confirmed hubs in Chapter 4 at the firewall and secure web gateway. Set a 30-day review.
- Put the two dynamic telecom listeners on a 24-hour watchlist and keep shared CDN addresses off automated block lists.
Within 72 hours
- Restrict workstation and server egress to ports 8787, 8790, 18787, 8888 and 2000 outside documented development subnets.
- Deploy Suricata SIDs 9528001 to 9528008. Send SID 9528004 (open-directory index) to a hunting queue, not an inline drop.
- Schedule the Sentinel query and the three Sigma rules. Name any sanctioned RStudio (8787) or Jupyter (8888) servers in the allowlists by exact host.
Within 7 to 30 days
- Pivot across proxy and NDR logs on JARM
3fd3fd16d...8956(368 ARTEX endpoints) and21d14d000...5d(204 endpoints) on PEG TECH, YANCY and CNServer ranges. - Ask for a second signal (URI, certificate or process) before blocking any suspected or candidate address.
- Add the pass rules your environment needs, such as internal RStudio servers, before enabling Suricata inline blocking.
8. Detection Rules
The full rules are in the Detection rules tab. 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. This report includes no captured ARTEX or DarkSword traffic, so test the HTTP strings in your own environment before blocking on them.
| Rule file | What it catches |
|---|---|
odin_artex_and_darksword_clusters.rules | Suricata 7: eight signatures for ARTEX and DarkSword check-ins, tasking responses, module requests, panel requests and open-directory indexes |
sentinel_artex_8787_and_darksword_cluster_hunt.kql | Microsoft Sentinel and Defender XDR: contact with the 19 confirmed addresses, and non-browser traffic to the cluster ports joined with process command lines |
sigma_artex_darksword_implant_and_beaconing.yml | Sigma (3 rules, Linux process creation): ARTEX listener or agent processes, DarkSword worker processes, and certificate or firewall provisioning scripts for the cluster ports |
odin_artex_darksword_cluster_implants.yar | YARA (4 rules): ARTEX listener and controller configs, DarkSword controller configs, and panel or staging manifests |
How the Suricata signatures work
| SID | Looks for | Ports |
|---|---|---|
| 9528001 | Outbound POST to /api/artex/ with "node_uuid" and "beacon_int" in the body | 8787, 8790, 18787, 7878 |
| 9528002 | Inbound 200 response with "artex_task" and "jitter_ms" in the body | From 8787, 8790, 18787 |
| 9528003 | Request to /artex/v1/sync with an X-Artex-Session: header | 80, 443, 8443 |
| 9528004 | Directory listing (Index of /) that contains artex_; policy alert for hunting | From 8787, 8081, 8443, 50549 |
| 9528005 | POST to /ds/v2/session with an X-DarkSword-ID: header | 443, 8443, 80 |
| 9528006 | POST to /darksword/queue with "task_opcode" in the body | 8888, 18888, 8088 |
| 9528007 | Request to /ds/module/pull with a user agent that starts DarkSword-Node/ | 2000, 22000 |
| 9528008 | Request to /api/syndicate/health with an X-Syndicate-Token: header | 3000, 8080, 8787, 8888 |
Rules on port 443 or 8443 see the HTTP content only where TLS is decrypted. Where it is not, pair the rules with the JARM values in Chapter 5 and the Sentinel query.
How the other rules work
The KQL query looks back 7 days. It flags any contact with the 19 confirmed addresses, with no exclusion on that branch. It also flags non-browser traffic to ten cluster ports, excluding browsers by exact path and scanners by exact device name. It joins these events with command lines that name artex, darksword, the rule URIs or the cluster ports. Port-only cases need 12 or more connections.
The first Sigma rule needs an ARTEX or Python process with /api/artex/, artex_relay or a --listen-port 8787 or 8790 argument. The second flags command lines with darksword or the DarkSword URIs. The third flags openssl req or iptables commands naming the cluster ports. All three exclude only named lab hosts.
| YARA rule | Fires when (file under 512 KB) |
|---|---|
ODIN_Artex_8787_8790_Relay_Listener_Config | ARTEX_RELAY or artex_task, plus 8787 or 8790, plus /api/artex/ |
ODIN_Artex_TLS_Controller_Session_Manifest | /artex/v1/sync and X-Artex-Session, plus beacon_int or jitter_ms |
ODIN_DarkSword_MultiPort_Controller_Config | A matching pair: the session header and path, the queue path and task_opcode, or the module path and user agent |
ODIN_Regional_C2_Cohosted_Staging_Panel_Manifest | The panel health path and token header, or artex_staging_root and cohosted_opendir |
9. Indicators
The Indicators tab holds 20 published indicators: 20 confirmed C2 endpoints (IP and port, on 19 distinct addresses). The 2 JARM fingerprints appear in this report only. It holds no file hashes. Six watchlist endpoints from the same selection are withheld, because they are suspected or candidate hosts.
| Type | Count | Where |
|---|---|---|
| Confirmed C2 endpoints (IP and port) | 20 | Indicators tab |
| JARM fingerprints for hunting | 2 | This report only |
| Watchlist endpoints | 6 | Withheld |
| Additional confirmed DarkSword hub addresses | 6 | Chapter 4 text only |
The full list of 516 confirmed hosts is not published here. Certificate names, the cross-family hosts and all watchlist addresses are withheld.
Confirmed endpoints in the Indicators tab
| Endpoint | Family | Provider (country) | Seen |
|---|---|---|---|
| 38.12.72[.]83:443 | ARTEX | PEG TECH INC (US) | 14 Nov 2026 |
| 38.38.175[.]199:443 | ARTEX | PEG TECH INC (US) | 14 Nov 2026 |
| 103.48.132[.]2:443 | ARTEX | YANCY LIMITED (US) | 14 Nov 2026 |
| 156.234.57[.]93:443 | ARTEX | YANCY LIMITED (HK) | 14 Nov 2026 |
| 8.216.54[.]254:443 | ARTEX | Alibaba US Technology (JP) | 18 Nov 2026 |
| 8.166.116[.]93:80 | ARTEX | Hangzhou Alibaba (CN) | 18 Nov 2026 |
| 8.137.13[.]182:8888 | ARTEX | Hangzhou Alibaba (CN) | 13 Nov 2026 |
| 64.90.1[.]153:8083 | ARTEX | NetLab Global (HK) | 13 Nov 2026 |
| 45.77.24[.]71:8443 | ARTEX | The Constant Company (JP) | 13 Nov 2026 |
| 110.42.238[.]147:7676 | ARTEX | Shenzhen Tencent (CN) | 13 Nov 2026 |
| 103.45.68[.]143:80 | ARTEX | CTG Server Limited (HK) | 14 Nov 2026 |
| 172.245.215[.]3:443 | ARTEX | HostPapa (US) | 13 Nov 2026 |
| 103.43.71[.]178:2000 | DarkSword | CyberForest LLC. (KR) | 13 Nov 2026 |
| 27.124.7[.]56:443 | DarkSword | CTG Server Limited (HK) | 13 Nov 2026 |
| 27.124.7[.]56:8443 | DarkSword | CTG Server Limited (HK) | 13 Nov 2026 |
| 193.218.201[.]146:443 | DarkSword | Spartan Host Ltd (US) | 13 Nov 2026 |
| 206.187.214[.]37:443 | DarkSword | cognetcloud INC (HK) | 13 Nov 2026 |
| 43.229.112[.]254:443 | DarkSword | YANCY LIMITED (US) | 13 Nov 2026 |
| 85.137.62[.]160:443 | DarkSword | SOLLUTIUM EU (NL) | 13 Nov 2026 |
| 86.38.247[.]228:443 | DarkSword | Datacamp Limited (SG) | 13 Nov 2026 |
45.77.24[.]71 and 172.245.215[.]3 present self-signed certificates that carry their own IP address as the subject.
JARM fingerprints for hunting (text only)
| JARM | Use |
|---|---|
3fd3fd16d3fd3fd21c42d42d00000078e356c38fb85055d97a8bed930c8956 | ARTEX on PEG TECH INC, 368 endpoints (173 confirmed) |
21d14d00021d21d00042d43d00000091f9827a8676a9d9f27d421962a09b5d | Mostly DarkSword on CNServer LLC, 204 endpoints (183 DarkSword, 21 ARTEX) |
10. Scope and Limits
- Server side only. The data shows listeners, certificates and TLS fingerprints on internet-facing servers. It shows no victim activity and no proof that any implant reached a target.
- Family labels come from upstream. "ARTEX" and "DarkSword" are labels attached to the records. "DarkSword" in particular is a feed label, not attribution to any named group or vendor report.
- Clusters are a model. Shared certificates and JARM values link hosts inside a cluster. Only two candidate hosts link the families, so this report makes no claim that one group runs both.
- Most endpoints are unconfirmed. Of the 2,674 endpoints, 2,148 are suspected, candidate or held for review. Port 8787, 8888 and 2000 also host benign services.
- Rule testing. Every rule file loads without errors in its engine (Suricata 7.0.12, YARA-X, pySigma, Kusto.Language). The rules have not been replayed against live traffic. Their HTTP strings are not drawn from captured traffic in this report.
11. Sources
- MITRE ATT&CK, Non-Standard Port (T1571): attack.mitre.org/techniques/T1571 (verified on 20 November 2026).
- MITRE ATT&CK, Application Layer Protocol: Web Protocols (T1071.001): attack.mitre.org/techniques/T1071/001 (verified on 20 November 2026).
- MITRE ATT&CK, Acquire Infrastructure: Virtual Private Server (T1583.003): attack.mitre.org/techniques/T1583/003 (verified on 20 November 2026).
- Salesforce Engineering, Easily Identify Malicious Servers on the Internet with JARM: engineering.salesforce.com (verified on 20 November 2026).
- Open Information Security Foundation (OISF), Suricata 7.0 User Guide: HTTP Keywords: docs.suricata.io (verified on 20 November 2026).
What is…?
Short explanations of the tools and families named in this study.
What is ARTEX?
ARTEX is an autonomous red-team console that chains AI agents and security tools. Operators fork and rebrand it; its default listener port is 8787.
More about ARTEXFeed labelWhat is DarkSword?
DarkSword is a family label from an upstream threat feed for multi-port TLS and HTTP C2 servers. It names a fingerprint group, not an actor.
More about DarkSwordTechniqueWhat is JARM?
JARM is an open TLS server fingerprinting method from Salesforce. It hashes how a server answers a set of crafted TLS handshakes.
More about JARMFrequently asked questions
What are ARTEX and DarkSword?
ARTEX is the family label for ARTEX, an autonomous red-team console that Odin covered in Monographs 04 and 11. Its default console port is 8787. In this study it makes up 2,312 endpoints on 2,157 hosts. Of these, 498 are confirmed controllers, mostly on ports 443 and 80.
How do the operators scale?
They rent many servers from the same provider and stamp them from one template. YANCY LIMITED hosts 237 confirmed ARTEX controllers, and PEG TECH INC hosts 175. Together that is 82.7% of all confirmed ARTEX controllers.
Why not block all 2,674?
Because most of them are not proven, and some are shared. Only 526 endpoints (19.7%) are confirmed. The other 2,148 include several kinds of address that a blunt block would get wrong.
What can defenders hunt for?
The templates that make these fleets cheap also make them visible. The Detection rules tab holds Suricata signatures, a Microsoft Sentinel query, three Sigma rules and four YARA rules. 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 Indicators tab holds 20 confirmed endpoints; the 2 JARM values for hunting are in the full report.