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.

In biology, mutation rate separates viruses from mammals. Fast mutation = evasion. Slow mutation = dominance. Zero mutation = either supreme confidence or supreme laziness.

The Outlaw payload hasn't changed a single byte in 71 days.

The Immutable Payload

SHA256 a8460f446be540410004b1a8db4083773fa46f7fe76fa84219c93daa1669f8f2 was first observed on March 31, 2026. It was last observed on June 10, 2026. In between: 85 downloads to 81 unique IP addresses. Same hash. Same binary. Not one byte different.

In an industry obsessed with polymorphic malware, metamorphic engines, and runtime-generated payloads, the Outlaw botnet deploys the same static binary for 2.5 months. This isn't naivety โ€” it's a calculated bet: the targets are unmonitored systems where hash-based detection doesn't exist.

Weekly Deployment Cadence

WeekDownloadsUnique IPsFailure Rate
Mar 3077~50%
Apr 61010~40%
Apr 1366~45%
Apr 2044~50%
Apr 2777~45%
May 455~50%
May 111414~35%
May 182825~40%
May 2511โ€”
Jun 122โ€”
Jun 811โ€”

The pattern tells a story. Steady 5-10 infections per week (background propagation), then a May 11-18 explosion โ€” 42 downloads in two weeks from 39 unique IPs. Then near-silence.

The spike isn't random. It's a coordinated push: someone acquired new credential lists, new scanning infrastructure, or simply decided to scale. The crash afterward suggests either operational disruption, goal achieved, or depleted target space.

The 40% Failure Rate

During the May spike: 42 successful a8460f44 downloads alongside 29 empty-hash failures. A 40% failure rate at peak load.

This is the C2 architecture cracking under pressure. The single distribution server โ€” identified in TI-2026-052C as lacking CDN redundancy โ€” can serve baseline traffic. But when the scanning infrastructure finds 39 new targets in two weeks, the download server intermittently fails.

The operators could fix this with a $5/month CDN. They don't. Either they don't care about 40% loss, or the infrastructure is intentionally minimal to reduce forensic exposure. Each additional server is another IP to burn, another hosting account to maintain, another potential seized box.

The Polymorphic Counterpoint

While Outlaw deploys one immutable binary, a parallel operation takes the opposite approach:

The "sshd" Replacements โ€” 24 Unique Binaries

PeriodUnique HashesPattern
Mar 24-3111Rapid experimentation โ€” new binary every day
Apr 1-158Continued polymorphism, slight slowdown
Apr 16-303Settling on proven variants
May-June2Near-silence or undetected

22 of 24 "sshd" hashes appear exactly once, from exactly one IP. Each attacker compiles a unique binary โ€” different enough to evade hash-based signatures, similar enough to serve the same purpose: replacing the SSH daemon with a backdoored version.

This is the sophisticated end of the spectrum. Where Outlaw says "we don't care about detection," the sshd backdoor operators say "we must never match a known signature." Two philosophies of persistence, operating in the same victim space.

Ecosystem Consolidation

The hash diversity timeline reveals an ecological pattern:

WeekUnique HashesTotal DownloadsDiversity Index
Mar 2311120.92 (high diversity)
Mar 3013350.37 (moderate)
Apr 68270.30 (declining)
May 43100.30 (low)
May 185520.10 (monoculture)
Jun 22111.0 (single species)

In the first week, every download was a different binary โ€” 11 unique hashes in 12 downloads. By May, a single payload dominates 80%+ of all activity. The ecosystem collapsed into a monoculture.

This mirrors biological competitive exclusion: when one organism (Outlaw/a8460f44) is sufficiently aggressive at eliminating competition (the kill -9 secure.sh command from TI-2026-052C), other species can't maintain presence. The malware literally kills competing miners on each infected host.

The result: a homogeneous botnet. One binary rules. Others are either killed on arrival or pushed to different ecological niches (the sshd backdoors persist because they serve a different function โ€” access rather than mining).

The Exception: 94f2e4d8

One secondary hash defied the monoculture: 94f2e4d8d44368747fe11fa207dfb6094e55b1f5e0cc4b1c32b9e3c6e7b8e2fd. Twelve downloads from twelve unique IPs, spanning the full 91-day observation window (March 30 โ€” June 29).

Also named "sshd." Also never mutated. But deployed to 12 IPs versus Outlaw's 81. This is either:

  • The same operator's secondary payload for specific target profiles
  • A competing operator with comparable infrastructure stability but smaller scale
  • A specialized variant (e.g., ARM vs x86, or rooted vs non-rooted persistence)

Its survival alongside a8460f44 suggests the two don't compete โ€” they coexist. Perhaps because they target different CPU architectures or implement different persistence mechanisms. The Outlaw botnet has historically supported multiple architectures from a single C2, making interpretation 1 most likely.

The Source Infrastructure

Download Sources Identified

SourceDownloadsUnique HashesActive Period
sshd (SFTP transfer)3925Mar 24 โ†’ Jun 30
raw.githubusercontent.com81Apr 5 (single day)
31.170.22.205 (Latvia)52Jun 13 (single day)
"notwork-monitoring"44Jun 8 (single day)
202.155.10.11221Mar 27

The dominant delivery method is SFTP transfer named "sshd" โ€” the malware copies itself through the same SSH session it used for access. No external HTTP download, no second connection. The compromised SSH session IS the delivery channel.

GitHub as C2: 8 downloads from raw.githubusercontent.com in a single day (April 5), all from one IP. This actor stored their payload in a GitHub repository โ€” abusing the platform's CDN for malware distribution. Fast, reliable, and extremely difficult to pre-emptively block.

31.170.22.205 (Latvia): The "Whisper" infrastructure from TI-2026-052B. Active for exactly one day โ€” deploy, infect, rotate.

"notwork-monitoring": A creative typosquat of "network-monitoring" โ€” designed to appear legitimate in process listings. Four different hashes in one day = rapid iteration testing.

Read Between the Lines

The mutation rate is an operational signature more revealing than the malware itself.

Zero mutation (Outlaw): The operators don't fear detection. Their targets โ€” compromised IoT devices, unmonitored VPS instances, residential connections โ€” don't run antivirus, don't submit to sandboxes, don't check hashes against threat feeds. When your victims can't see you, why bother changing?

Maximum mutation (sshd backdoors): These operators fear detection intensely. Each binary unique. Each deployment singular. Their targets are worth more โ€” monitored servers, infrastructure that someone might actually forensically examine.

The ecosystem collapse: Over 91 days, diversity decreased from 11 unique families to 1 dominant species. In biology, monocultures are vulnerable โ€” one immunity mechanism kills everything. In cybercrime, monoculture means a single signature could theoretically disrupt 81 nodes simultaneously. But only if someone deploys that signature on unmonitored residential IoT in Indonesia, Nigeria, and Brazil.

The operators know their victims better than we do. Their mutation rate is calibrated not to the threat landscape, but to the monitoring landscape of their specific target demographic. And that demographic doesn't monitor.

The Thread Forward

In TI-2026-052E, we synthesize the complete malware supply chain: developers who compile, distributors who host, operators who deploy, and customers who profit. The mutation rate reveals who occupies which role โ€” and who is paying whom.

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€” 4 / 5 Next โ†’