Skip to content
NewCybercrime Report 2026 · Eight Chapters · A Post Is Not An Incident → Explore The ChaptersCybercrime Report 2026 → Explore
All full reports
Full reportCommodity malwareMonograph 51 / 5519 min read

Loader and Stealer Feeds: Seven Relay Candidates, 908 Held Rows

How 3,500 examined endpoints across 3,311 hosts split into seven cloud and VPS relay candidates, 700 victim-labelled Ramnit rows, and 208 telecom and CDN collisions

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

1. Executive Summary

In short

In a two-week observation window ending 21 November 2026, Odin by Brandefense examined 3,500 endpoints across 3,311 hosts from 11,819 sightings labelled as Emotet, RedLine Stealer, Raccoon Stealer, LokiBot, and Ramnit. Only seven cloud and VPS endpoints across six IPv4 addresses survived initial collision checks, and none carried independent C2 beacon corroboration. Meanwhile, all 700 examined Ramnit rows carried an upstream victim label and 208 rows sat on broadband ISPs or shared CDNs. Keep unverified feed IPs off static block lists and hunt HTTP, SOAP, and process signals.

Security teams frequently import bulk malware feeds into perimeter firewalls without cyber threat intelligence platform review, expecting every listed IPv4 address to belong to an attacker server. Across these feeds, upstream exports mix cloud and VPS listings with broadband, shared proxy, and victim-labelled addresses.

Blocking raw Emotet or Ramnit IP lists cuts off broadband subscribers, breaks websites behind Cloudflare, and leaves proxy-fronted panels that defenders can review with external attack surface management online. Because none of the 3,311 hosts in our study spans two families, we present the five families as an analytical model.

1

Only seven cloud and VPS relay candidates remain among 3,500 examined rows.

Across 11,819 global sightings and 34 upstream "confirmed" records, collision filtering left five Emotet endpoints on four cloud and VPS addresses plus two RedLine Stealer endpoints. Because independent C2 beacon corroboration is not documented for those six addresses, all 3,500 examined endpoints across 3,311 hosts stay withheld.

2

All 700 examined Ramnit rows carry an upstream victim label and were withheld.

Ramnit accounts for 6,804 global sightings (57.6% of the five-family total) and carries an explicit victim classification in upstream feeds. That pool includes four entries marked verified on FTP port 21 and port 8443. Blocking raw Ramnit exports blackholes 599 victim-labelled enterprise and broadband hosts in China, South Korea, Thailand, and Indonesia rather than verified attacker command servers.

3

Telecom carriers and shared CDNs account for 208 collateral collision rows.

Seventy-three LokiBot candidates sit behind shared Cloudflare proxies; 66 Emotet rows sit on broadband carriers or Akamai edge nodes; 65 RedLine and four Raccoon rows belong to subscriber ISPs. Network signatures for SOAP contracts, PHP gates, and multipart POSTs can flag matching traffic without static IP blocks; each match needs triage.

Three Decisions for Leaders

  1. Require an independent protocol or endpoint signal before blocking feed IPv4 addresses. Even the seven dedicated cloud and VPS relay candidates across Amazon Web Services, DigitalOcean, and Hetzner lacked documented C2 beacon proof. Keep all 3,500 uncorroborated, ISP, CDN, and victim-labelled endpoints out of static firewall deny lists.
  2. Quarantine victim-tagged botnet feeds and shared CDN ranges before firewall import. Route Ramnit victim records, broadband telecom networks, and Cloudflare or Akamai proxy addresses to threat-hunting workflows instead of outbound deny lists.
  3. Shift loader and stealer defense to wire-level HTTP, SOAP, and endpoint process rules. Alert on RedLine WCF SOAP headers, LokiBot PHP submission paths, Raccoon /config posts, and regsvr32.exe or rundll32.exe outbound network sessions.

The Study in Numbers

NumberWhat it is
11,819Global feed sightings across Emotet, RedLine Stealer, Raccoon Stealer, LokiBot, and Ramnit
3,500 / 3,311Examined endpoints (700 per family) across 3,311 distinct hosts and 2,368 /24 subnets
7 / 6 IPsDedicated cloud and VPS relay candidates (5 Emotet, 2 RedLine) across 6 withheld IPv4s; C2 uncorroborated
700 / 599Examined victim-labelled Ramnit endpoints withheld across 599 hosts
208 / 42Broadband telecom ISP, Emotet-labelled carrier, and Cloudflare or Akamai rows across 42 networks
2,585Other unconfirmed port, TLS fingerprint, and historical panel candidates held for hunting
8Published TLS JARM indicators in the Indicators tab (plus 5 text-only wire signatures in Chapter 9)

How to use this report.

Security leaders can start with Chapters 1, 6, and 7. Network and firewall teams should read Chapters 3, 4, 5, and 7. SOC analysts and detection engineers should go directly to Chapters 8 and 9. Threat intelligence analysts will find the hosting breakdown and scope limits in Chapters 2, 5, 6, and 10; the Odin Research blog collects related cases.

2. Five Malware Labels, One Analytical Model

In short

Zero hosts among the 3,311 distinct hosts in our sample carry labels from more than one family. We present Emotet, RedLine Stealer, Raccoon Stealer, LokiBot, and Ramnit together as an analytical model of feed-list collisions, not as one threat actor. Keep uncorroborated feed lists off firewalls and hunt protocol signals.

Modular Windows loaders (Emotet), Malware-as-a-Service information stealers (RedLine Stealer and Raccoon Stealer), PHP credential panels (LokiBot), and legacy banking worms (Ramnit) serve distinct criminal roles. Even so, threat feeds often ship all five inside one indicator stream without separating attacker servers from victim or proxy addresses. None of the five families in this report overlaps with Monographs 24 or 35.

Comparing host records across all 3,500 examined endpoints shows zero shared hosts across families. Our 3,311 distinct hosts equal 678 Emotet, 683 RedLine, 700 Raccoon, 651 LokiBot, and 599 Ramnit hosts, with zero shared custom certificates across the seven cloud and VPS relay candidates. At the /24 prefix level (2,368 distinct /24 blocks), 21 commercial hosting blocks hold candidates from two families.

Those 21 blocks include 10 Raccoon and RedLine, 8 LokiBot and RedLine, 2 LokiBot and Raccoon, and 1 Emotet and Raccoon prefixes. The shared prefixes may reflect co-tenant VPS hosting at M247 Europe SRL, Hetzner Online GmbH, OVH SAS, and Delis LLC; they do not establish operator identity. Figure 1 maps the four functional groups in our analytical model and their defensive split.

Figure 1: Victim botnet feeds and ISP relays dwarf seven cloud and VPS relay candidates
Figure 1: Victim botnet feeds and ISP relays dwarf seven cloud and VPS relay candidates

From 11,819 global sightings to 3,500 examined endpoints

Global sightings total 11,819 entries across the five families. Ramnit leads with 6,804 sightings across 47 countries (57.6%), followed by RedLine Stealer with 2,247 across 48 countries (19.0%) and Emotet with 1,053 across 59 countries (8.9%).

LokiBot accounts for 885 sightings across 8 countries (7.5%), and Raccoon Stealer accounts for 830 across 34 countries (7.0%). Upstream feeds assign Emotet, RedLine, and Raccoon to command-and-control, LokiBot to panel management, and Ramnit explicitly to a victim category.

Within our 3,500-endpoint sample (700 per family), 247 rows carry last-seen dates inside the two-week window ending 21 November 2026 (59 Emotet, 9 RedLine, and 179 Ramnit rows). The other 3,253 rows carry earlier upstream feed listing dates from 2020 through October 2026 rather than continuous observation by Odin by Brandefense.

Only seven cloud and VPS endpoints (0.20% of examined rows; 0.059% of global sightings) across six IPv4 addresses survived initial collision filtering, and none had independent C2 corroboration. All six addresses and the remaining 3,493 endpoints (99.80%) across 3,305 hosts stay withheld.

FamilyCategorySightingsFeed "verified"Examined / hostsCloud/VPS relay candidatesISP & CDN heldVictim heldOther candidates
EmotetC21,05326700 / 6785 (4 IPs held)660629
RedLine StealerC22,2474700 / 6832 (2 IPs held)650633
Raccoon StealerC28300700 / 700040696
LokiBotManagement8850700 / 651073 (Cloudflare)0627
RamnitVictim6,8044700 / 59900700 (100%)0

Geographic and hosting split

Across the 3,500 examined endpoints, 2,881 map to 65 countries, while 619 historical panel records carry an unattributed country code. Those 619 rows include 587 LokiBot, 14 Raccoon, 11 RedLine, 4 Ramnit, and 3 Emotet entries, with 583 lacking current ASN routing.

China leads attributed volume with 498 endpoints (438 victim-labelled Ramnit rows and 60 Emotet rows), followed by the United States (451), the Netherlands (407), Germany (304), and Russia (191). South Korea (176, including 161 victim-labelled Ramnit rows), the United Kingdom (97), Singapore (79), Finland (71), France (70), and Uruguay (57) complete the top locations.

Commercial cloud or VPS providerRowsPrimary families seenStudy note
Amazon.com, Inc.190Emotet (124), RedLine (45), Raccoon (21)2 upstream-confirmed relay candidates (US, UK) + 188 other candidates
DigitalOcean, LLC117Emotet (102), Raccoon (9), RedLine (6)4 upstream-confirmed Emotet endpoints (3 droplets) + 113 other candidates
Hetzner Online GmbH97Raccoon (47), Emotet (39), RedLine (11)1 upstream-confirmed RedLine endpoint (FI) + 96 other candidates
RootLayer Web Services Ltd.80RedLine Stealer (80)Includes 2 withheld port 3389 RDP collisions
OVH SAS70Emotet (35), RedLine (20), Raccoon (14), LokiBot (1)Shared VPS candidates (0 upstream-confirmed)
Partner LLC70Raccoon Stealer (70)Historical HTTP port 80 candidates
STARK INDUSTRIES SOLUTIONS LTD54Raccoon (51), RedLine (3)Offshore VPS candidates
M247 Europe SRL53Raccoon Stealer (53)European HTTP port 80 candidates
AEZA GROUP Ltd44Raccoon Stealer (44)Commercial HTTP port 80 candidates
SERV.HOST GROUP LTD41RedLine Stealer (41)High-port WCF VPS candidates

3. Cloud and VPS Relay Candidates: Emotet and RedLine Stealer

In short

Emotet and RedLine Stealer account for all seven dedicated cloud and VPS relay candidates (six IPv4 addresses) in this study. Twenty-three other upstream "confirmed" rows in those two families failed review due to telecom listings, lure certificates, or port 3389 RDP collisions. Because independent C2 beacon corroboration is not documented for the six cloud and VPS addresses, withhold them from static block lists and deploy wire rules for Emotet POSTs and RedLine SOAP contracts.

Emotet (MITRE ATT&CK S0367) places infected Windows computers (T1583.005) in front of backend cloud servers as Epoch reverse proxies on ports 443, 8443, 8080, 2083, 80, and 8010. On infected hosts, Emotet commonly launches DLLs via regsvr32.exe (T1218.010) and rundll32.exe (T1218.011).

RedLine Stealer (S1240), built in C# .NET, exposes Windows Communication Foundation (WCF) BasicHttpBinding SOAP services (hxxp://tempuri[.]org/IRemoteEndpoint/) over high TCP ports such as 1912, 55615, 25920, 27667, 49157, and 5800. Figure 2 maps all five families to their wire behaviors and detections.

Figure 2: SOAP contracts, PHP gates, and multipart POSTs map to six wire rules
Figure 2: SOAP contracts, PHP gates, and multipart POSTs map to six wire rules

Emotet: five cloud and VPS relay candidates vs broadband proxy listings

Emotet accounts for 1,053 sightings across 59 countries, with 700 endpoints examined across 678 hosts (21 multi-port hosts across 43 rows) and 622 /24 subnets. Port 443 leads with 563 endpoints (80.4%), followed by 8443 (48), 2083 (29), 8080 (27), 80 (19), 8010 (2), and 444 (2). Leading countries include the United States (171), Germany (117), China (60), Russia (45), France (41), Singapore (37), the United Kingdom (22), and the Netherlands (20).

Twenty-six examined Emotet rows carried an upstream "confirmed" label. Independent C2 corroboration for the five cloud and VPS endpoints is not documented.

  • US AWS port 443 relay candidate (address withheld) on Amazon.com, Inc. in the United States, presenting a custom domain TLS certificate (withheld) and JARM prefix 29d29d07d29d.
  • German DigitalOcean ports 80 and 8010 relay candidate (address withheld) on DigitalOcean, LLC in Germany, a dual-port droplet listening on HTTP ports 80 and 8010.
  • Two US DigitalOcean port 80 relay candidates (addresses withheld) on DigitalOcean, LLC in the United States, paired droplets sharing one custom certificate profile on port 80.

Twenty-one other upstream-confirmed Emotet rows failed collision review. Two port 443 hosts sat on a regional broadband carrier in southern China, where infection and proxy behavior were not verified for each host. Six dual-port 443 and 8443 rows across three Huawei Cloud hosts presented spoofed financial-brand certificates without corroborated Emotet C2 traffic.

Six Alibaba Cloud port 443 rows came from a single uncorroborated batch. Seven HTTP port 80 and 8010 rows at JSC Selectel and Data Storage Center JSC also lacked handshake proof. Overall, 66 Emotet-labelled rows (9.4%) sit on telecom carriers (45 rows across 16 countries) or Akamai Connected Cloud edge nodes (21 rows). Those telecom rows include four port 8443 nodes in South Korea sharing JARM prefix 2ad2ad0002ad.

RedLine Stealer: two dedicated relay candidates vs port 3389 RDP and telecom collisions

RedLine Stealer accounts for 2,247 sightings across 48 countries, with 700 endpoints examined across 683 hosts and 427 /24 subnets. High WCF ports dominate: TCP 1912 (124, 17.7%), 80 (112, 16.0%), 55615 (107, 15.3%), 25920 (57, 8.1%), 443 (35, 5.0%), 27667 (21), 55123 (8), and 16383 (8). Sixty-five RedLine rows expose default Windows machine certificates.

Four RedLine endpoints carried an upstream "confirmed" label. Independent C2 corroboration for the two dedicated cloud and VPS endpoints is not documented:

  • One UK AWS port 49157 endpoint and one Finnish Hetzner port 5800 endpoint (addresses withheld) on Amazon.com, Inc. and Hetzner Online GmbH lacked independent WCF SOAP beacon corroboration.
  • Two Dutch port 3389 endpoints (addresses withheld) on RootLayer Web Services Ltd. (alongside one suspected port 3389 host) were withheld because Windows Remote Desktop TLS handshakes triggered false C2 labels without WCF SOAP evidence.

Sixty-five RedLine rows (9.3%) belong to subscriber telecom networks, led by 57 port 25920 leases on a national telecom carrier in Uruguay plus eight broadband leases across seven other countries.

4. Proxy-Fronted Panels and Victim Feeds: Raccoon, LokiBot, and Ramnit

In short

Raccoon Stealer (700 examined), LokiBot (700 examined), and Ramnit (700 examined from 6,804 sightings) produced zero corroborated block-list addresses. Seventy-three LokiBot rows point to shared Cloudflare proxy IPs, Raccoon rows carry stale 2025 listing dates, and all 700 Ramnit rows carry an upstream victim label. Catch LokiBot and Raccoon via HTTP URI signatures and keep Ramnit victim lists off firewalls.

Raccoon Stealer and LokiBot: stale port 80 VPS pools and 73 Cloudflare proxies

Raccoon Stealer (MITRE ATT&CK S1148) has 830 sightings across 34 countries. We examined 700 endpoints on 700 hosts and 419 /24 subnets. Its HTTP POST check-in uses /config with bot_id and config_id.

In our sample, 685 of 700 Raccoon rows (97.9%) listen on TCP port 80. Leading locations include the Netherlands (211), Germany (119), the United States (77), Russia (76), Finland (54), and the United Kingdom (43). Zero Raccoon rows carried upstream verification, and all 700 carry feed listing dates on or before 15 December 2025 (696 candidates plus 4 telecom ISP rows). Blocking recycled VPS addresses nearly a year later risks breaking reassigned tenant workloads.

Raccoon Stealer VPS providerExamined rows
Partner LLC70
M247 Europe SRL53
STARK INDUSTRIES SOLUTIONS LTD51
Hetzner Online GmbH47
AEZA GROUP Ltd44
Galaxy LLC40

LokiBot (MITRE ATT&CK S0447; 885 sightings across 8 countries) accounts for 700 examined endpoints across 651 hosts and 489 /24 subnets, including 46 multi-port hosts across 95 rows. Its servers run PHP administration panels (PvqDq929BSx_A_D_M_I_N_a.php) and binary credential gates (fre.php) with Charon; Inferno User-Agent strings on port 80 (570, 81.4%) and port 443 (121, 17.3%).

Zero LokiBot rows carried upstream verification, and all 700 carry listing dates on or before 7 May 2026. Critically, 73 LokiBot endpoints (10.4%) resolve to Cloudflare, Inc. shared anycast IPv4 addresses. Blocking those 73 Cloudflare proxy IPs breaks enterprise access to thousands of unrelated websites while leaving the origin PHP panel online.

Ramnit: why all 700 victim-labelled rows (from 6,804 sightings) are withheld

Ramnit accounts for 6,804 global sightings across 47 countries (57.6% of all 11,819 sightings) and carries an explicit victim classification in upstream feeds. Following the 2015 Europol takedown across roughly 3.2 million infected PCs, upstream feeds continue to log Ramnit victim or sinkhole-peer sightings, though infection or sinkhole status was not independently verified on every host.

Our 700 examined Ramnit endpoints span 599 distinct hosts and 432 /24 subnets, including 55 multi-port hosts exposing up to nine ports on a single host.

Exposed ports on those 700 victim-labelled Ramnit rows reflect ordinary customer services. Leading ports include HTTP port 80 (257), 82 (43), 8088 (39), 8081 (28), 88 (26), 443 (26), 81 (26), 9201 (21), and FTP port 21. Four Ramnit rows carried a "confirmed" tag in the upstream feed. Those four rows comprised three FTP port 21 servers in China and one port 8443 telecom host in Thailand with an OAuth single-sign-on lure certificate.

Overall, 655 of 700 Ramnit rows (93.6%) sit in China (438), South Korea (161, with 158 on one national carrier), Thailand (34), and Indonesia (22). Under our victim-protection rules, all 700 Ramnit rows (100%) are withheld.

5. Why Blind IP Blocking Breaks Broadband ISPs, CDNs, and Victim Networks

In short

Combining 700 victim-labelled Ramnit endpoints with 208 telecom and CDN collision endpoints means 908 of our 3,500 examined rows (25.9%) sit on victim-labelled customer hosts, residential ISPs, or shared Cloudflare and Akamai proxies. Windows TLS fingerprints also collide directly between RedLine candidates and victim-labelled Ramnit rows. Replace blind IP feed blocks with URI, SOAP, and process detections.

Figure 3 shows how victim-labelled rows, broadband listings, and Cloudflare-fronted PHP gates enter raw threat feeds. It contrasts them with seven cloud and VPS candidates across six withheld IPv4 addresses.

Figure 3: Why raw Emotet and Ramnit IP feeds blackhole innocent networks
Figure 3: Why raw Emotet and Ramnit IP feeds blackhole innocent networks

Three ways raw loader and botnet feeds cause self-inflicted outages

  1. Victim-tagged worm and sinkhole logs (700 Ramnit endpoints on 599 hosts). Upstream victim tags on customer FTP (port 21) and web devices (ports 80, 81, 82, 88, 8081, 8088) flag suspected victim or sinkhole-peer hosts without verifying every host, not attacker C2 infrastructure.
  2. Broadband proxy candidates and DHCP rotation (114 telecom ISP rows across 40 carriers). Subtracting 73 Cloudflare and 21 Akamai rows from the 208-row collision pool (42 networks) leaves 114 telecom ISP rows. Those rows include 57 RedLine port 25920 leases in Uruguay, 45 Emotet-labelled telecom leases across 16 countries, 8 other RedLine ISP leases, and 4 Raccoon ISP leases. Infection and proxy behavior were not verified for each host, and dynamic DHCP rotation can reassign residential addresses within days.
  3. Shared CDN reverse proxies and edge nodes (94 Cloudflare and Akamai endpoints). Blocking 73 Cloudflare proxy addresses or 21 Akamai Connected Cloud addresses causes immediate business web outages while attackers rotate domains. Figure 4 contrasts the six dedicated cloud and VPS relay candidate hosts (seven endpoints, addresses withheld) with the 3,305 other withheld hosts.
Figure 4: Six cloud and VPS relay candidate hosts contrast with 3,305 other withheld hosts
Figure 4: Six cloud and VPS relay candidate hosts contrast with 3,305 other withheld hosts

TLS JARM overlap and certificate collisions across 1,163 endpoints

TLS JARM fingerprints appear on 1,163 examined endpoints across 369 unique hashes (2,337 non-JARM rows), and X.509 certificates appear on 1,066 endpoints across 697 subjects (2,434 non-TLS rows). Emotet's top JARM (2ad2ad0002ad prefix, 96 rows) spans both cloud candidates and broadband telecom rows in South Korea and China.

Worse still, the next three top JARMs reflect standard Windows SChannel TLS stacks that collide directly between RedLine candidates and victim-labelled Ramnit hosts. Prefix 14d14d16d14d spans 55 rows (34 RedLine, 13 Ramnit, 8 others) and 26d26d16d26d spans 47 rows (31 Ramnit, 14 RedLine, 2 others). Prefix 2ad2ad16d2ad spans another 43 rows (31 RedLine, 11 Ramnit, 1 Raccoon).

Certificate or JARM groupRowsFamily breakdownRelay candidatesDefender guidance
Custom domain certificates837259 Emotet, 215 Ramnit, 197 RedLine, 155 Raccoon, 11 LokiBot4Withhold unverified IPs and domains; pair with HTTP/TLS wire rules
Default Windows host certs8665 RedLine, 16 Raccoon, 5 Ramnit1Never block port 3389 RDP certs; pair with WCF SOAP rules
Generic portal mimic certs4418 Emotet, 18 Raccoon, 8 RedLine0Hunt only; reused across VPS parking templates
Financial-brand lure certs2322 Emotet (Huawei Cloud), 1 Raccoon0Withhold IPs; 6 feed-confirmed Emotet rows lacked C2 proof
OAuth and SSO lure certs1918 Ramnit, 1 Emotet0Withhold victim hosts; monitor outbound credential phishing
IP-literal (26) & localhost (19)45Seen across all 5 families0Generic server defaults; never block on certificate alone
Top 3 Windows SChannel JARMs14579 RedLine, 55 Ramnit-labelled, 8 Raccoon, 2 Emotet, 1 LokiBot0Heavy victim and Windows Server overlap; pair with EDR logs
Emotet 2ad2ad0002ad JARM9696 Emotet (cloud and broadband telecom rows)0Hunt paired with regsvr32.exe and rundll32.exe egress

6. Feed Labels Against Cloud and VPS Relay Candidates

In short

Importing raw lists for these five families straight into a firewall would block 599 victim-labelled Ramnit hosts, 114 broadband ISP addresses, and 94 Cloudflare or Akamai edge addresses. Only seven dedicated cloud and VPS endpoints (0.20% of examined rows; 0.059% of global sightings) survived collision checks, and all six IPv4 addresses remain withheld pending independent C2 corroboration. Keep raw feed lists off firewalls and hunt wire and endpoint signals.

Figure 5 compares global sighting volumes and the 208-row ISP and CDN pool against the seven cloud and VPS relay candidates.

Figure 5: Seven cloud and VPS relay candidates across 11,819 global sightings
Figure 5: Seven cloud and VPS relay candidates across 11,819 global sightings
Group in our analysisFeed "confirmed"Feed "suspected"Feed "candidate"Total rowsHostsShareAction
Cloud and VPS relay candidates700760.20%Withhold IP; hunt with wire and endpoint rules
Other unverified candidates21302,5342,5852,49873.86%Withhold; hunt with wire and endpoint rules
Subscriber ISPs and shared CDNs252012082085.94%Withhold; protect broadband users and Cloudflare
Victim-labelled Ramnit rows41568170059920.00%Withhold; never block victim-classified feeds

Why 27 of 34 feed-confirmed rows failed collision checks

Twenty-seven of the 34 upstream "confirmed" records (79.4%) failed initial collision review. Four were victim-labelled Ramnit hosts on port 21 (three in China) and port 8443 (one in Thailand), and two were Emotet-labelled broadband telecom rows in southern China.

Two more were Dutch port 3389 Remote Desktop collisions at RootLayer Web Services Ltd., while six Huawei Cloud port 443 and 8443 financial-lure rows lacked Emotet C2 traffic. The final 13 rows included six Alibaba Cloud port 443 rows from one uncorroborated batch. Seven HTTP port 80 and 8010 rows at JSC Selectel and Data Storage Center JSC also lacked handshake verification.

Four other ways to read the 3,493 non-relay endpoints

  1. Pre-existing services on victim-labelled hosts (700 Ramnit rows on 599 hosts). Some victim-labelled hosts expose up to nine ports, including 21, 80, 81, 82, 88, 8081, 8088, and 9201. The ports may serve FTP, cameras, or web traffic. Individual infection or sinkhole status was not verified.
  2. Dynamic broadband NAT and telecom ISP listings (114 telecom ISP rows). Subscriber-network rows (57 RedLine port 25920 leases in Uruguay and 45 Emotet-labelled telecom leases) sit on residential carrier ranges where individual infection or proxy status was not verified.
  3. Shared CDN anycast proxies and edge caches (94 Cloudflare and Akamai rows). Operators place 73 LokiBot PHP panels behind Cloudflare because blocking shared anycast IPs breaks legitimate websites.
  4. Reassigned VPS addresses, Windows SChannel overlaps, and RDP servers (2,585 candidates). Older Raccoon (2025) and LokiBot (May 2026) VPS listings, shared Windows SChannel JARMs, and port 3389 servers require endpoint or payload corroboration.

7. What Defenders Should Do

Perimeter firewall and network security

  • Do not import raw Emotet, RedLine, Raccoon, LokiBot, or Ramnit feed IPv4 addresses into static firewall block lists without an independent protocol, beacon, or endpoint signal.
  • Keep Cloudflare, Akamai Connected Cloud, residential telecom ISP ranges, and victim-classified Ramnit feeds out of automated firewall deny lists.

SOC and endpoint detection

  • Deploy the six Suricata 7 signatures in Chapter 8 on perimeter and SSL-inspection links. Those rules alert on RedLine IRemoteEndpoint SOAP envelopes, LokiBot /PvqDq929BSx_A_D_M_I_N_a.php and /fre.php posts, Emotet bare-IPv4 multipart POSTs, and Raccoon /config check-ins behind Cloudflare or ISP NAT.
  • Run the Sentinel KQL hunt and two Sigma rules across Windows endpoint logs. Isolate hosts when regsvr32.exe or rundll32.exe connects to public Emotet ports (80, 443, 2083, 8010, 8080, 8443) or .NET and script utilities open RedLine high ports (1912, 5800, 16383, 25920, 27667, 49157, 55123, 55615).
  • Scan unpacked binaries, sandbox drops, and memory dumps with the three YARA rules for RedLine WCF contracts, LokiBot Charon; Inferno gates, and Raccoon /config parameters.

Threat intelligence management

  • Strip victim-categorized botnet feeds (Ramnit), standalone Windows SChannel JARMs, and port 3389 Remote Desktop collisions out of firewall External Dynamic Lists.

8. Detection Rules

The full rules are in the Detection rules tab. The Suricata file passes a static syntax check with zero problems. The rules were not loaded on live Suricata, Sentinel Kusto, Sigma, or YARA engines for this edition, so test and tune them before enabling automated isolation.

Rule fileWhat it catches
odin_emotet_redline_raccoon_and_lokibot_c2.rulesSuricata 7 (6 rules): RedLine WCF SOAP headers and bodies, LokiBot PHP panel and fre.php posts, Emotet bare-IPv4 multipart POSTs, and Raccoon /config check-ins
sentinel_loader_stealer_c2_and_victim_noise_hunt.kqlMicrosoft Sentinel and Defender XDR: regsvr32.exe/rundll32.exe Emotet port egress and .NET or script binaries hitting RedLine high ports
sigma_loader_stealer_relays_and_wcf_beacons.ymlSigma (2 Windows network rules): regsvr32.exe/rundll32.exe reaching Emotet ports, and .NET or script binaries reaching RedLine WCF ports
odin_emotet_redline_raccoon_and_lokibot_artifacts.yarYARA (3 rules): RedLine IRemoteEndpoint WCF configs, LokiBot Charon; Inferno and fre.php stagers, and Raccoon /config check-ins

How the Suricata, KQL, Sigma, and YARA rules work

SID or RuleLooks forAction
SID 9551001HTTP POST with SOAPAction IRemoteEndpoint headerRedLine WCF SOAP header
SID 9551002XML SOAP envelope calling CheckConnect, ScanResult, or GetArgumentsRedLine WCF SOAP body
SID 9551003HTTP URI /PvqDq929BSx_A_D_M_I_N_a.phpLokiBot PHP admin panel
SID 9551004HTTP POST to /fre.php with Charon; Inferno User-AgentLokiBot credential gate
SID 9551005Multipart POST to bare IPv4 Host without Referer or OriginEmotet HTTP beacon (hunt)
SID 9551006HTTP POST to /config containing bot_id and config_idRaccoon config check-in
Sigma Rule 1Exact regsvr32.exe or rundll32.exe reaching public Emotet portsHigh (contain loader)
Sigma Rule 2Exact .NET or script host reaching RedLine WCF high portsHigh (contain stealer)
YARA Rules 1 to 3Bounded (<2 MB) RedLine WCF, LokiBot Charon, and Raccoon /config payloadsHunt / Contain

The Sentinel KQL query looks back seven days across DeviceNetworkEvents and DeviceProcessEvents. It correlates outbound regsvr32.exe and rundll32.exe connections on Emotet ports. It also checks .NET or script connections on RedLine WCF ports, excluding exact lab hostnames in ApprovedRedTeamHosts.

TacticTechniqueStudy observationDetection
Resource DevelopmentT1583.003 Virtual Private ServerCloud and VPS relay candidates across DigitalOcean, Hetzner, and AWS8 JARM indicators; Sigma 1, 2; Sentinel KQL
Resource DevelopmentT1583.005 Botnet66 Emotet-labelled ISP/edge rows and 700 Ramnit victim-labelled nodesFeed quarantine; Sigma 1
Defense EvasionT1218.010 / T1218.011 Regsvr32 and Rundll32Emotet DLL loader network activitySigma 1; Sentinel KQL
Command and ControlT1071.001 Web ProtocolsRedLine SOAP, LokiBot /fre.php, Raccoon /config, Emotet POSTSIDs 9551001 to 9551006; YARA 1 to 3
Command and ControlT1573.002 Asymmetric CryptographyEmotet port 443 and RedLine port 49157 TLS relay candidates8 JARM indicators; Sentinel KQL
Credential AccessT1555.003 Credentials from Web BrowsersRedLine, Raccoon, and LokiBot browser password theftSIDs 9551001 to 9551004, 9551006; Sigma 2; YARA 1 to 3

9. Indicators

The Indicators tab holds 8 published indicators: 8 TLS JARM fingerprints used for hunting. All six cloud and VPS IPv4 addresses remain withheld because no independent C2 beacon is documented. They cover seven address-and-port entries; one German DigitalOcean droplet had ports 80 and 8010.

The other 3,493 examined endpoints are also withheld. There are no file hashes in this report. Five text-only HTTP and SOAP wire signatures also appear below.

TypeCountWhere to use it
TLS JARM fingerprints (Indicators tab)8TLS hunting paired with endpoint process logs; never block on a JARM alone
HTTP and SOAP wire signatures (text-only)5Suricata and HTTP proxy rules (sid:9551001 to 9551006)
Withheld cloud and VPS relay candidates (6 unique IPv4s)7Withheld pending independent protocol or beacon signal; do not block

JARM fingerprints and wire signatures to hunt

IndicatorFamilyContext
29d29d07d29d29d21c29d29d29d29dce2e4c8c53174ecbf5529ce7584d5518EmotetUpstream-confirmed US AWS port 443 endpoint (address withheld)
2ad2ad0002ad2ad0002ad2ad2ad2ade1a3c0d7ca6ad8388057924be83dfc6aEmotetSeen across 96 cloud and broadband telecom rows
29d29d00029d29d21c29d29d29d29d8da29ade5a975f32a8d3f61299c69dc4EmotetShared across four Alibaba Cloud port 443 candidate endpoints
2ad2ad16d2ad2ad00042d42d0000008fe5654c9239cdb4052d3ab65a579afaEmotetHuawei Cloud port 443 financial-lure certificate cluster
14d14d16d14d14d08c14d14d14d14dfd9c9d14e4f4f67f94f0359f8b28f532RedLine StealerWindows SChannel JARM on 55 rows including 13 victim-labelled Ramnit rows
26d26d16d26d26d22c26d26d26d26dfd9c9d14e4f4f67f94f0359f8b28f532RamnitWindows JARM on 47 rows including 31 victim-labelled Ramnit rows and 14 RedLine rows
2ad2ad16d2ad2ad22c2ad2ad2ad2adfd9c9d14e4f4f67f94f0359f8b28f532RedLine StealerWindows SChannel JARM on 43 rows including 11 victim-labelled Ramnit rows
29d3fd00029d29d21c42d43d00041df61f6b5fa6973e9fa14c8155669c6633Raccoon StealerLinux Nginx JARM across 21 RedLine, 16 Raccoon, and 1 LokiBot candidates
hxxp://tempuri[.]org/IRemoteEndpoint/RedLine StealerText-only WCF SOAPAction header contract; SIDs 9551001, 9551002
/PvqDq929BSx_A_D_M_I_N_a.phpLokiBotText-only PHP C2 administration panel URI; SID 9551003
/fre.phpLokiBotText-only Charon binary credential submission gate; SID 9551004
/configRaccoon StealerText-only HTTP POST check-in path (bot_id, config_id); SID 9551006
Content-Type: multipart/form-dataEmotetText-only bare-IPv4 HTTP POST without Referer; SID 9551005

We withhold all 3,500 examined endpoints across 3,311 hosts. These include 7 uncorroborated relay rows on 6 hosts, 700 Ramnit rows on 599 hosts, and 208 ISP/CDN rows on 208 hosts. The other 2,585 candidates span 2,498 hosts; unverified certificate domains and Windows machine hostnames are also withheld.

10. Scope and Limits

  • Observation window. Findings cover a two-week observation window ending 21 November 2026. Earlier dates on 3,253 examined records (including Raccoon Stealer through December 2025 and LokiBot through May 2026) are upstream feed listing dates, not continuous monitoring by Odin by Brandefense.
  • Feed labels are not attribution. Malware family names and verification tags come from upstream feeds. Because independent protocol or beacon corroboration is not documented for the seven cloud and VPS relay candidates (six IPv4 addresses), all 3,500 examined endpoints remain withheld from public block lists.
  • Analytical model. Zero hosts span two families across all 3,311 hosts. Figures 1 and 3 present an analytical model of loader, stealer, and victim-feed collisions rather than a unified threat group.
  • No victim or subscriber hosts contacted. Odin by Brandefense did not interact with Ramnit victim-labelled machines or consumer broadband routers, and all 3,311 examined hosts remain withheld.
  • Rule validation. Detection rules were verified statically and were not loaded on live production engines for this edition.

11. Sources

  1. MITRE ATT&CK. S0367 Emotet. https://attack.mitre.org/software/S0367/ (verified 21 November 2026)
  2. MITRE ATT&CK. S1240 RedLine Stealer. https://attack.mitre.org/software/S1240/ (verified 21 November 2026)
  3. MITRE ATT&CK. S1148 Raccoon Stealer. https://attack.mitre.org/software/S1148/ (verified 21 November 2026)
  4. MITRE ATT&CK. S0447 Lokibot. https://attack.mitre.org/software/S0447/ (verified 21 November 2026)
  5. CISA. Alert AA20-280A: Emotet Malware. https://www.cisa.gov/news-events/cybersecurity-advisories/aa20-280a (verified 21 November 2026)
  6. CISA. Alert AA20-266A: LokiBot Malware. https://www.cisa.gov/news-events/cybersecurity-advisories/aa20-266a (verified 21 November 2026)
  7. Europol EC3. World's most dangerous malware EMOTET disrupted through global action. https://www.europol.europa.eu/media-press/newsroom/news/world%E2%80%99s-most-dangerous-malware-emotet-disrupted-through-global-action (verified 21 November 2026)
  8. Europol EC3. Botnet [Ramnit] taken down through international law enforcement cooperation. https://www.europol.europa.eu/media-press/newsroom/news/botnet-taken-down-through-international-law-enforcement-cooperation (verified 21 November 2026)
  9. MITRE ATT&CK. T1583.005 Acquire Infrastructure: Botnet. https://attack.mitre.org/techniques/T1583/005/ (verified 21 November 2026)
  10. Salesforce Engineering. JARM: active TLS server fingerprinting. https://github.com/salesforce/jarm (verified 21 November 2026)

What is…?

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

All explanations

Frequently asked questions

Why do Ramnit and Emotet feeds blackhole innocent networks?

More than half of all sightings in our study (6,804 of 11,819, or 57.6%) carry a Ramnit label under an explicit victim category. Across 700 examined Ramnit endpoints on 599 hosts, exposed ports belong to ordinary FTP, web, and router services. These include ports 21, 80, 81, 82, 88, 8081, and 8088.

Why do even the six remaining cloud and VPS relays require corroboration?

Reviewing all 34 upstream "confirmed" records eliminated 27 rows (79.4%) on collision grounds. Those 27 rows included four victim-labelled Ramnit rows, two broadband Emotet listings, two Dutch port 3389 Remote Desktop collisions, six Huawei Cloud financial-lure rows, and 13 single-batch cloud rows.

How do you catch proxy-hidden stealers without blocking Cloudflare or ISPs?

Raccoon Stealer (700 examined rows, 97.9% on HTTP port 80, listed on or before December 2025) produced zero corroborated block-list servers. LokiBot (700 rows on ports 80 and 443, listed through May 2026) also produced zero corroborated block-list servers. Three Windows TLS JARM hashes also occur across 79 RedLine rows and 55 Ramnit victim-labelled rows. Instead of blocking recycled VPS or CDN addresses, inspect protocol payloads and process behavior:

Are these five families operated by one group?

No evidence links them to one operator. Zero hosts among all 3,311 distinct hosts carry labels from two families, and the 21 shared /24 blocks may reflect co-tenant VPS hosting. We present the five families as an analytical model to help teams separate durable C2 signals from victim and ISP noise.

Defender kit

Indicators of compromise

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

IDTypeIndicatorPortContextAction
R51-IOC-01 JARM 29d29d07d29d29d21c29d29d29d29dce2e4c8c53174ecbf5529ce7584d5518 443 emotet · TLS fingerprint on upstream-confirmed US AWS port 443 endpoint (address withheld) Hunt
R51-IOC-02 JARM 2ad2ad0002ad2ad0002ad2ad2ad2ade1a3c0d7ca6ad8388057924be83dfc6a 443 emotet · Dominant Emotet TLS fingerprint across 96 cloud and broadband telecom rows Hunt
R51-IOC-03 JARM 29d29d00029d29d21c29d29d29d29d8da29ade5a975f32a8d3f61299c69dc4 443 emotet · TLS fingerprint shared across four Alibaba Cloud port 443 candidate endpoints Hunt
R51-IOC-04 JARM 2ad2ad16d2ad2ad00042d42d0000008fe5654c9239cdb4052d3ab65a579afa 443 emotet · TLS fingerprint on Huawei Cloud port 443 financial-lure certificate cluster Hunt
R51-IOC-05 JARM 14d14d16d14d14d08c14d14d14d14dfd9c9d14e4f4f67f94f0359f8b28f532 443 redlinestealer · Windows SChannel TLS fingerprint across 55 endpoints including 13 Ramnit victim-labelled rows Hunt
R51-IOC-06 JARM 26d26d16d26d26d22c26d26d26d26dfd9c9d14e4f4f67f94f0359f8b28f532 8443 ramnit · Windows TLS fingerprint across 47 endpoints including 31 Ramnit victim-labelled rows and 14 RedLine candidates Hunt
R51-IOC-07 JARM 2ad2ad16d2ad2ad22c2ad2ad2ad2adfd9c9d14e4f4f67f94f0359f8b28f532 443 redlinestealer · Windows SChannel TLS fingerprint across 43 endpoints including 11 Ramnit victim-labelled rows Hunt
R51-IOC-08 JARM 29d3fd00029d29d21c42d43d00041df61f6b5fa6973e9fa14c8155669c6633 443 raccoonstealer · Linux Nginx TLS fingerprint shared across 21 RedLine, 16 Raccoon, and 1 LokiBot candidates Hunt

28 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.

YARA odin_emotet_redline_raccoon_and_lokibot_artifacts.yar 64 lines
Download
Suricata odin_emotet_redline_raccoon_and_lokibot_c2.rules 18 lines
Download
Sentinel KQL sentinel_loader_stealer_c2_and_victim_noise_hunt.kql 63 lines
Download
Sigma sigma_loader_stealer_relays_and_wcf_beacons.yml 101 lines
Download