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 29 / 1915 min read

Mobile Overlay C2s, Cloud Worms and Recon-Phishing Fleets

How HookBot, P2PInfect, EvilGoPhish and dual-use recon suites split into block and watch decisions

By
Odin by Brandefense
Published
19 November 2026
Reading time
15 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 mapped 717 endpoints on 687 hosts across ten infrastructure families. Android banking panels, cloud worms, reverse-proxy phishing kits and recon suites sit side by side. Recon suites make up 93.0% of the endpoints. Only one HookBot controller, 152.70.24[.]49, is corroborated in the window as adversary infrastructure. Block it, keep everything else on watchlists, and close public Redis access.

This report covers four kinds of internet-facing infrastructure. Android banking trojan panels steal credentials and control phones. Cloud worms spread through exposed Redis and SSH services. Phishing kits proxy login pages to steal session cookies. Recon suites map an organization's external attack surface. The same tools serve attackers and defenders, so the key question for each group is whether to block or to watch.

1

One HookBot controller is the only address published for blocking.

152.70.24[.]49 on Oracle Cloud in Germany is a HookBot command-and-control (C2) server seen in the window. Eleven more HookBot addresses are older feed entries, kept on watchlists. Five Octopus panels on adjacent IBM Cloud addresses fit a lab and go to review.

2

P2PInfect peers and EvilGoPhish servers stay on watchlists, not block lists.

Sixteen P2PInfect nodes, 15 on Alibaba Cloud, use high gossip ports from 60103 to 60140. These are most likely infected third-party servers. Nine EvilGoPhish servers pair a GoPhish admin port with a reverse proxy, but authorized phishing tests use the same kit.

3

Dual-use recon suites dominate the volume and stay on watch.

ARL (488 endpoints), ScopeSentry (138) and reNgine (41) make up 667 endpoints. Security teams run these suites too, so none is blocked. Separately, one exposed workbench in Japan held 1,725 files, including AI agent logs and an authentication archive.

Three Decisions for Leaders

  1. Block the one confirmed controller and watch everything else. Put 152.70.24[.]49 on perimeter and mobile VPN block lists. Route the older HookBot entries, the Octopus cluster, the 667 recon endpoints and all worm and phishing nodes to time-bound watchlists.
  2. Require phishing-resistant sign-in and protect mobile apps. FIDO2 passkeys defeat reverse-proxy session theft. Android restricted settings and runtime integrity checks limit the Accessibility abuse that HookBot, Daam and L3MON depend on.
  3. Close public Redis and audit open file servers. Turn on Redis protected mode and authentication, and keep it off the internet. Remove public directory listings from engineering and automation servers.

The Study in Numbers

NumberWhat it is
717Endpoints across ten families, on 687 distinct hosts
1HookBot controller corroborated in the window and published for blocking
16Older HookBot entries (11) and Octopus panels (5) kept on watchlists, addresses withheld
667ARL, ScopeSentry and reNgine recon endpoints, all kept on watchlists (93.0%)
16P2PInfect peer nodes; 15 sit on Alibaba Cloud
9EvilGoPhish phishing servers; 4 use the default admin port 3333
1,725Files on one exposed AI agent workbench in Japan
24Hosts that bind several services, including 2 that run both ARL and ScopeSentry

How to use this report.

Security leaders can read Chapter 1 and Chapter 7. Detection engineers should start with Chapters 7 to 10. CTI analysts will find the infrastructure detail in Chapters 2 to 6.

2. Four Layers of Infrastructure

In short

The 717 endpoints fall into four functional layers: recon, phishing, worms and mobile C2. Recon suites are large and noisy; mobile panels and worm meshes are small and compact.

Defenders who ingest raw internet scan feeds face an imbalance. Recon frameworks produce a large volume of banners on ports such as 5003 and 8082. Mobile banking panels and worm meshes run in small, fast-changing clusters. Treating both alike either floods analysts with false positives or blocks benign security tools.

We group the ten families into four layers. This grouping is an analytical model. Nothing observed links the operators of one layer to another.

LayerFamiliesEndpoints
ReconARL, ScopeSentry, reNgine667
Initial access phishingEvilGoPhish9
Cloud and edge wormsP2PInfect, Moobot18
Mobile C2HookBot, Octopus, Daam, L3MON23
Figure 1: Four operational layers split 717 endpoints into block and watchlist actions
Figure 1: Four operational layers split 717 endpoints into block and watchlist actions

Counts by family

Each endpoint is an IP and port pair. The 717 endpoints sit on 687 hosts, because two servers run two families each.

FamilyRoleEndpointsHostsBlockWatch
ARLDual-use recon4884740488
ScopeSentryDual-use recon1381360138
reNgineDual-use recon4129041
P2PInfectCloud peer-to-peer worm1616016
HookBotAndroid overlay C21313112
EvilGoPhishReverse-proxy phishing9909
OctopusC2 panel5505
DaamAndroid overlay and RAT3303
L3MONAndroid NodeJS RAT2202
MoobotEdge botnet worm2202

Where the endpoints run

Six countries lead.

CountryEndpointsShare
China33246.3%
United States16422.9%
Hong Kong8111.3%
Germany304.2%
Singapore253.5%
Japan142.0%

The layers use different hosting. ARL runs mostly in China (284), the United States (128) and Hong Kong (52). ScopeSentry shows a similar spread: China 43, Hong Kong 28 and the United States 20. The largest recon hosting networks are Shenzhen Tencent (155 rows), SonderCloud (91) and Hangzhou Alibaba (82).

The HookBot and Octopus panels sit on European and global VPS providers instead. IBM Cloud hosts five in Germany, Contabo two in Germany, Hostinger two, and Oracle one in Germany. P2PInfect keeps 12 of its 16 nodes (75.0%) on Alibaba (US) Technology in Singapore.

3. Android Banking Panels

In short

HookBot, Octopus, Daam and L3MON account for 23 endpoints. Only one HookBot controller is corroborated in the window; the other 22 endpoints stay on watchlists.

Modern Android banking trojans combine three tricks. A WebView overlay draws a fake login screen over a banking app. Android AccessibilityService lets the malware read the screen and tap buttons. Real-time VNC screen streaming lets the operator move money directly on the phone.

Figure 2: Distinct port bindings and TLS fingerprints separate mobile C2s, worms and recon suites
Figure 2: Distinct port bindings and TLS fingerprints separate mobile C2s, worms and recon suites

HookBot

HookBot has 13 endpoints. One controller, 152.70.24[.]49 on port 80, is corroborated in the window. It sits on Oracle Corporation in Germany and presents a certificate for a custom domain, which is withheld.

Eleven more HookBot addresses come from older feed entries. None of them has a record inside this window. Cloud addresses get reassigned, so these entries go to watchlists and their addresses are withheld. The thirteenth endpoint listens on port 2223 and stays a watchlist candidate.

The older entries still carry useful fingerprints. Eight listen on port 80 and four on port 443 across the 12 controller records. Five present the stock TRAEFIK DEFAULT CERT certificate, which points to a containerized deployment behind the Traefik reverse proxy.

FingerprintRecordsHosting
TRAEFIK DEFAULT CERT, JARM prefix 3fd3fd...3Contabo GmbH (DE, 2), RouterHosting LLC (US)
TRAEFIK DEFAULT CERT, JARM prefix 26d26d...1The Constant Company (CL)
TRAEFIK DEFAULT CERT, JARM prefix 29d29d...1Amazon.com (ID)
JARM prefix 0003fd...1Hostinger (LT)

Hunt the certificate subject together with the JARM values. Treat a match as a lead, not as proof of a live controller.

Octopus

Five Octopus panels listen on port 80 inside adjacent IBM Cloud subnets in Frankfurt, Germany. Four form a contiguous block and the fifth sits in a nearby subnet. All five were seen on 13 November 2026.

Octopus is also the name of an open-source red-team C2 framework. Five adjacent cloud addresses fit a test lab as well as an attacker. We therefore withhold the addresses and send the cluster to analyst review rather than to block lists.

Daam and L3MON

Daam and L3MON are not corroborated yet, so they stay on watchlists. Their addresses are withheld.

FamilyEndpointsPortsHosting
Daam3443Saudi Arabia, United States and India; one US node presents CN=Plesk on IONOS
L3MON222533, 22222Amazon (US) and Iliad Italia (IT)

L3MON is an open-source NodeJS remote-management panel for Android devices. Its default port 22533 is a useful hunting signal by itself. Both L3MON candidates were seen on 19 November 2026.

4. P2PInfect and Moobot Cloud Worms

In short

P2PInfect builds a peer-to-peer mesh out of exposed Redis and SSH servers. Its 16 nodes are most likely infected third-party machines, so this report treats them as victims and publishes no addresses.

P2PInfect does not rely on one central server. It scans for exposed Redis instances and SSH servers, then turns infected hosts into peers that talk over high ports. Unit 42 describes it as a Rust-based, self-replicating worm. Taking down one node does not stop the mesh.

What the P2PInfect nodes look like

FeatureCount
Nodes in total16
On Alibaba Cloud15 (93.8%): 12 in Singapore, 2 in China, 1 in Hong Kong
Gossip ports 60103 to 6014012
Port 60882
Port 82811
Port 96841
Self-signed CN=gateway (shared JARM prefix 3fd3fd...)2
Self-signed CN=localhost3

Several nodes were seen on 19 November 2026, on ports 60103, 60136, 60139, 60140 and 6088.

Why no addresses are published

An infected cloud server belongs to someone who is likely unaware of the infection. Blocking it would hurt its owner as much as the worm. Publishing it would expose the owner. We therefore describe the nodes by port, certificate and hosting type only.

Moobot

Two Moobot nodes serve HTTP on port 80. Moobot is a Mirai variant that stages payloads over plain HTTP. One node sits on DA International Group in Bulgaria. Both stay on watchlists with addresses withheld.

5. EvilGoPhish and the Recon Suites

In short

Nine EvilGoPhish servers combine a GoPhish campaign panel with a reverse proxy that captures session cookies. They sit next to 667 recon endpoints that map login portals, subdomains and certificates.

Recon suites and phishing kits serve adjacent stages of an attack. Recon finds the login portals. The phishing kit then copies one of them. Figure 3 shows this sequence as a conceptual model, not as one observed intrusion.

Figure 3: Surface discovery, AiTM session capture and mobile overlays target distinct layers
Figure 3: Surface discovery, AiTM session capture and mobile overlays target distinct layers

EvilGoPhish

EvilGoPhish joins two tools. GoPhish runs the email campaign through an admin API on port 3333. An Evilginx-style reverse proxy on port 443 relays the real login page. It captures the session cookie and the one-time MFA code as the victim signs in.

FeatureDetail
Servers9, one per host
CountriesGermany 3, United States 2, Ireland 1, China 1, India 1, Indonesia 1
Ports3333 (4 servers), plus 3322, 3334, 33335, 10002 and 443
Shared TLS fingerprint3 servers share the Go crypto/tls JARM 40d40d40d00000000043d43d00043d6aff5ab0f4fc31897186312c9e639319
Lookalike certificates2 DigitalOcean servers present certificates for lookalike domains (withheld)
Activity8 were seen on 19 November 2026; two 3333 servers, on IONOS (DE) and Amazon (IE), were first seen that day

The Go JARM value is common to many Go programs, so it is a hunting lead, not proof by itself. Pair it with port 3333 or a lookalike certificate.

Companies also run EvilGoPhish for authorized phishing exercises. A single port or JARM match cannot tell the two apart. All nine servers stay on watchlists until a lure domain confirms hostile use. Their addresses and the targeted brands are withheld.

ARL, ScopeSentry and reNgine

These three open-source attack-surface management (ASM) frameworks enumerate subdomains, ports and web assets. They make up 93.0% of the endpoints (667 of 717).

SuiteEndpointsHostsMain portsFingerprints
ARL4884745003 (340, 69.7%), 443 (96, 19.7%)123 nodes share JARM 15d3fd16d29d29d00042d43d000000fbc10435df141b3459e26f69e76d5947; 19 suspected hosts also expose RDP on 3389 with self-signed certificates named after internal 10.x addresses
ScopeSentry1381368082 (119, 86.2%), 18082 (3), 8443 (3)Hosted in China 43, Hong Kong 28, United States 20 and Japan 11
reNgine41298082 (18), 443 (14), 80 (4), 8000 (2)17 present the default CN=rengine.example[.]com; 15 of those share JARM 2ad2ad0002ad2ad0002ad2ad2ad2ada5fc7ec76b7ed0b839211317d3844486

reNgine has the most ports per host in this study, at 1.41 endpoints per host. One DigitalOcean stack in Germany runs it on ports 80, 443 and 8000 together.

Two Chinese servers run ARL and ScopeSentry side by side. One, on China Telecom, pairs ARL on 5003 with ScopeSentry on 8082. The other, on CHINANET BACKBONE, pairs ARL on 20000 and 43047 with ScopeSentry on 8082. Their addresses are withheld as dual-use.

6. An Exposed Agent Workbench

In short

One server in Japan exposed an open directory of 1,725 files. It mixed AI agent logs, an agent backup, a Gemini authentication archive and cloud deployment notes.

Operators sometimes leave a Python file server or a reverse proxy open while they stage tools. The directory tree then shows tool pairings that a banner alone cannot. We do not know who owns this server. It may belong to a developer rather than an attacker, so its address and file paths are withheld.

Figure 4: Exposed agent workbenches and co-hosted recon stacks show multi-tool consolidation
Figure 4: Exposed agent workbenches and co-hosted recon stacks show multi-tool consolidation

What the workbench held

The directory listing is dated 19 November 2026. Its main files are listed here.

ArtifactSizeWhat it is
Hermes agent backup archive821.1 MBComplete Hermes agent workspace and state
Bun runtime binary89.0 MBSelf-contained JavaScript and TypeScript agent runtime
Repository archive (name withheld)14.7 MBPackaged code repository in the web root
Gemini authentication archive12.8 MBCompressed Gemini API or CLI authentication bundle
Linux utility binary (name withheld)6.59 MBPrecompiled amd64 tool
Flask service log3.59 MBLog of a Python Flask orchestration service
OpenClaw agent log, dated 19 November3.43 MBSession log of the OpenClaw autonomous agent
OpenClaw agent log, dated 20 November292.5 KBThe next day's log, from the server's local UTC+9 clock
Headless browser log82.1 KBA Chrome DevTools Protocol daemon, Headless Browser v0.1.9
Cloud deployment notes100.4 KBNotes on deploying Cloudflare Workers, D1, KV and R2 with wrangler to a third-party domain (withheld)

The workbench also held a Hermes repository index, a backed-up proxy configuration and Cloudflare login logs. Together these show one person running autonomous agents, browser automation and cloud deployment from one machine. Model authentication files sat in an internet-readable folder.

Open directories on recon nodes

Open directories also show up inside the ARL fleet. Six ARL candidate endpoints on five hosts also serve an open HTTP file index. Three of the six also expose configuration or credential files next to the console.

Hosting (country)PortsNotes
Shenzhen Tencent (CN)5003, 5030Two ports on one host
Shenzhen Tencent (CN)5003Exposed configuration files
LUCIDACLOUD (HK)5003Exposed configuration files
LUCIDACLOUD (HK)5003Exposed configuration files
Tencent Building (KR)892Open directory

These hosts stay on watchlists for analyst review. Their addresses are withheld.

7. Block or Watch: Telling Hostile from Dual-Use

In short

Most endpoints here run tools that defenders also use. Blocking them all would break legitimate scanners and still miss rotating worm peers, so only one corroborated controller goes on block lists.

Three innocent explanations apply to large parts of this data:

  • Sanctioned security tools. Enterprise EASM and penetration-testing teams run ARL, ScopeSentry and reNgine.
  • Authorized phishing exercises. Security-awareness teams deploy EvilGoPhish on cloud servers.
  • Victim machines and shared addresses. P2PInfect peers are infected third-party servers. Some recon nodes sit on telecom subscriber networks where addresses rotate.

The table below shows how the 717 endpoints split into actions.

GroupEndpointsSuggested action
HookBot controller corroborated in the window1Block; review the lease after 30 days
Older HookBot feed entries11Watch; re-check before any block
Octopus panels that fit a test lab5Analyst review
Confirmed recon suites: 94 ARL, 23 reNgine, 10 ScopeSentry127Watch for 7 days; never block automatically
Recon nodes on telecom subscriber networks (4 confirmed, 1 suspected)5Watch for 24 hours
ARL candidates with an open directory6Watch for 7 days; analyst review first
Suspected recon nodes: 23 ARL, 2 reNgine25Watch for 7 days
Uncorroborated candidates, including the 16 P2PInfect peers and 9 EvilGoPhish servers537Watch for 72 hours; act only on corroboration

The telecom subscriber networks are China Unicom, China Telecom, ChinaTelecom Jiangsu, China Mobile and Telekomunikasi Indonesia. Addresses there change hands, so a long block would hit later, unrelated users.

The action bar at the foot of Figure 1 shows the same split.

8. What Defenders Should Do

Within 24 hours

  • Block the confirmed HookBot controller 152.70.24[.]49 on perimeter firewalls and mobile VPN gateways. Review the cloud lease after 30 days, because cloud addresses get reassigned.
  • Put the recon, worm and phishing groups on time-bound watchlists, as in Chapter 7. Do not block them automatically.

Within 72 hours

  • Take cloud Redis (6379) off the public internet. Turn on protected mode and ACL authentication.
  • Require SSH key-only access on cloud servers, so P2PInfect cannot recruit them.
  • Require FIDO2 or WebAuthn sign-in on workforce SSO portals. These methods defeat reverse-proxy session theft.
  • Watch new TLS certificates for lookalikes of your own domains.

Within 7 to 30 days

  • Enforce Android restricted settings and OWASP MASVS runtime integrity checks in mobile banking apps. This limits sideloaded Accessibility overlays and VNC screen capture.
  • Deploy the Suricata rules and the Sentinel hunt. Keep the ARL, ScopeSentry and reNgine rules in alert mode, never in drop mode.
  • Audit engineering and red-team servers for open HTTP directories that expose agent logs, backups or cloud authentication files.
  • Hunt the certificate subjects and JARM values in this report across your external attack surface tooling.

9. Detection Rules

In short

Four rule files cover the mobile panels, cloud worms, EvilGoPhish and recon suites. They hold eight Suricata signatures, one Sentinel query, three Sigma rules and four YARA rules.

The full rules are in the Detection rules tab.

Rule fileWhat it catches
odin_hookbot_p2pinfect_evilgophish_recon.rulesSuricata 7: eight signatures for mobile C2 check-ins, P2PInfect and Moobot fetches, EvilGoPhish campaign calls and recon task APIs
sentinel_mobile_overlay_p2pinfect_and_evilgophish_hunt.kqlMicrosoft Sentinel and Defender XDR: contact with the confirmed HookBot controller, and fleet ports used by Redis, Node, Python or GoPhish processes
sigma_p2pinfect_redis_ssh_worm_and_evilgophish.ymlSigma (3 rules): Redis or SSH child processes with worm markers, EvilGoPhish launches, and unapproved recon workers on Linux
odin_hookbot_l3mon_p2pinfect_evilgophish_artifacts.yarYARA (4 rules): HookBot or Daam panel configs, L3MON manifests, P2PInfect peer configs and EvilGoPhish configs

How the Suricata signatures work

All eight signatures inspect outbound HTTP from inside your network.

SIDLooks forPorts
9529001POST to /api/hookbot/ with "accessibility" and "vnc_state" in the body80, 443, 2223
9529002POST to /api/octopus/ with "agent_uuid" in the body80
9529003POST to /client_login with "device_model" in the body (Daam or L3MON)443, 22222, 22533
9529004/p2p/peers request with an X-Peer-Node: header (P2PInfect)6088, 8281, 9684, 60103 to 60140
9529005/moobot. fetch with a Wget/ user agent80
9529006POST to /api/campaigns/ with an X-EvilGoPhish-Key: header443, 3322, 3333, 3334, 10002, 33335
9529007POST to /api/task/ with "task_tag" (ARL), as a policy event443, 5003, 5004, 5080, 8443
9529008POST to /api/scan/create with "scan_engine" (ScopeSentry or reNgine), as a policy event443, 8000, 8082, 8443, 18082

SIDs 9529007 and 9529008 use the policy-violation class. Keep them in alert mode, and add pass rules for your own named EASM scanners. Self-signed 443 and 601xx channels need TLS inspection or endpoint data to be seen.

The Suricata file loads in Suricata 7.0.12. The signatures have not been replayed against captured traffic from each family, so test them in alert mode first.

How the other rules work

The KQL query looks back 7 days and has two parts. The first flags any device that contacts the confirmed HookBot controller address. This branch always alerts, even from an approved host. The second flags fleet ports such as 3333, 6088, 22533 and 60103 when redis-server, python3, node, curl, wget or gophish opens them.

It then joins process events that mention p2pinfect, evilgophish, l3mon or recon workers, or slaveof on Redis. An approved-host list holds exact hostnames for your own scanners and phishing simulators. That list never hides a confirmed controller match.

The first Sigma rule needs a parent of redis-server or sshd and a command line with a worm port, p2pinfect or slaveof. The second flags command lines that name evilgophish, evilginx, gophish_admin or :3333. The third flags ARL, ScopeSentry or reNgine worker launches. Each rule has an exact-hostname exclusion for sanctioned servers.

Each YARA rule is bounded to files under 512 KB and needs at least two independent strings.

YARA ruleFires when
ODIN_HookBot_Daam_Overlay_Panel_ConfigA HookBot or Daam marker appears with accessibility_inject or vnc_state
ODIN_L3MON_NodeJS_Android_Panel_ManifestAn L3MON marker, port 22533 or 22222, and /client_login appear together
ODIN_P2PInfect_Cloud_Worm_Peer_ConfigX-Peer-Node and /p2p/peers appear with a gossip port or redis_peer_sync
ODIN_EvilGoPhish_Hybrid_AiTM_ConfigAn EvilGoPhish marker, an admin marker and evilginx_sub appear together

10. Indicators

The Indicators tab holds 1 published indicator: 1 address and no hashes. It is the HookBot controller 152.70.24[.]49 on port 80, on Oracle Corporation in Germany.

TypeCountWhere
Confirmed controller address1Indicators tab
Older HookBot feed entries11Described in Chapter 3; addresses withheld
Octopus panels that fit a test lab5Described in Chapter 3; addresses withheld
Other watchlist endpoints (HookBot candidate, Daam, L3MON, P2PInfect, Moobot, EvilGoPhish, recon suites)700Described in Chapters 3 to 5; addresses withheld
File hashes from the exposed workbench5Withheld

Addresses of older HookBot entries, the Octopus cluster, dual-use recon hosts, infected worm peers, phishing servers of unknown purpose and the workbench are withheld. So are the workbench file hashes, lookalike domains and targeted brands.

Supporting fingerprints: TRAEFIK DEFAULT CERT, CN=gateway, CN=rengine.example[.]com, and the JARM values named in Chapters 3 to 5.

11. Scope and Limits

  • Service data, not victim logs. This report describes internet-facing services. It does not show what any panel, worm or phishing kit did to a specific victim.
  • Older HookBot entries. For 11 of the 12 HookBot controller records, the latest record predates this window, so these addresses are withheld and kept on watchlists.
  • Octopus may be a lab. Five adjacent IBM Cloud addresses running a panel named after an open-source red-team C2 can be a test lab. Our data cannot rule that out.
  • Layers are a model. The four-layer grouping is analytical. No shared certificate, address or campaign token links the recon, phishing, worm and mobile groups.
  • Dual-use tools. ARL, ScopeSentry, reNgine and EvilGoPhish serve both defenders and attackers. Our data cannot tell which use applies to a given host.
  • Worm peers are victims. P2PInfect nodes are most likely infected servers. Their owners are not adversaries.
  • Workbench owner unknown. The exposed workbench may belong to a developer. Its contents show tooling, not hostile action.
  • Rule testing. 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.

12. Sources

  1. Android Developers, AccessibilityService: developer.android.com/guide/topics/ui/accessibility/service (verified on 20 November 2026).
  2. OWASP Foundation, Mobile Application Security Verification Standard (MASVS): mas.owasp.org/MASVS (verified on 20 November 2026).
  3. Palo Alto Networks Unit 42, P2PInfect: The Rusty Peer-to-Peer Self-Replicating Worm: unit42.paloaltonetworks.com/peer-to-peer-worm-p2pinfect (verified on 20 November 2026).
  4. Redis Documentation, Redis security: redis.io/docs/latest/operate/oss_and_stack/management/security (verified on 20 November 2026).
  5. CISA, Implementing Phishing-Resistant MFA (fact sheet): cisa.gov/sites/default/files/publications/fact-sheet-implementing-phishing-resistant-mfa-508c.pdf (verified on 20 November 2026).
  6. OISF, Suricata 7.0 User Guide: HTTP keywords: docs.suricata.io/en/suricata-7.0.0/rules/http-keywords.html (verified on 20 November 2026).
  7. Salesforce Engineering, Easily Identify Malicious Servers on the Internet with JARM: engineering.salesforce.com/easily-identify-malicious-servers-on-the-internet-with-jarm-e095edac525a (verified on 20 November 2026).

What is…?

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

All explanations

Frequently asked questions

What did we find?

The 717 endpoints fall into four layers. This grouping is our analytical model. Nothing observed links the people behind one layer to another. The recon layer is by far the largest, at 93.0% of all endpoints. The mobile layer is small but holds the one controller we publish for blocking.

Which servers should we block?

Block one address: 152.70.24[.]49, a HookBot controller on Oracle Cloud in Germany that we corroborated in the window. HookBot is an Android banking trojan. It draws fake login screens over banking apps, abuses Android Accessibility to tap buttons, and streams the phone's screen to the operator. The Indicators tab lists this address.

Why not block the worms and phishing kits?

P2PInfect is a self-spreading worm. It finds Redis databases and SSH servers left open on the internet, infects them, and links them into a peer-to-peer mesh on high ports from 60103 to 60140. We found 16 nodes, 15 of them on Alibaba Cloud.

What about the recon suites?

ARL (488 endpoints), ScopeSentry (138) and reNgine (41) map subdomains, ports and login portals. Attackers use that map to choose phishing targets. Security teams use the same tools to find their own exposure.

What can defenders hunt for?

Addresses change quickly. Fingerprints change slowly, so they make better hunting signals. The full report lists the details, and a few stand out. Inside your own network, the most useful signal is a server that starts talking on these ports when it never did before. A Redis server that opens a connection on port 60103 deserves a look. So does a workstation that reaches the confirmed HookBot controller.

Defender kit

Indicators of compromise

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

IDTypeIndicatorPortContextAction
R29-IOC-01 IPv4 152.70.24[.]49 80 hookbot Block

27 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_hookbot_l3mon_p2pinfect_evilgophish_artifacts.yar 75 lines
Download
Suricata odin_hookbot_p2pinfect_evilgophish_recon.rules 20 lines
Download
Sentinel KQL sentinel_mobile_overlay_p2pinfect_and_evilgophish_hunt.kql 41 lines
Download
Sigma sigma_p2pinfect_redis_ssh_worm_and_evilgophish.yml 83 lines
Download