Report: 5.8 Million Knocks: A Three-Year Honeypot Record of SSH and Telnet Attacks
Malware activity is huge in volume yet almost entirely unique. Some 5,575,924 file interactions produced 1,838,290 distinct files, while a handful of scripts recur in hundreds of thousands of attacks.
| Measure | Value | Note |
|---|---|---|
| File interactions | 5,575,924 | About 4.2 per active attack |
| File creations (command redirection) | 447,720 | 8.0% of interactions |
| File transfers (uploads and downloads) | 5,128,204 | 92.0% of interactions |
| Unique files | 1,838,290 | About 1.4 per active attack |
| Unique files created | 609 | Each recreated about 735 times on average |
| Unique files transferred | 1,837,681 | Each seen about 2.8 times on average |
| Distinct download hosts | 766 |
The per-attack figures assume that file activity occurs in active attacks (1,319,254), which matches the honeypot’s definitions. Top 10 files by number of active attacks they appear in
| Rank | SHA-256 (abbreviated) | Attacks | Share of attacks | Type | Activity | VirusTotal verdict |
|---|---|---|---|---|---|---|
| 1 | a8460f44…f8f2 |
347,073 | 26.3% | Created | Attack prep | Malicious |
| 2 | 01ba4719…546b |
48,091 | 3.7% | Create | Attack prep | Undetected* |
| 3 | 1b20a210…2057 |
37,189 | 2.8% | Created | Attack recon | Undetected |
| 4 | 207e9bcc…ed2a |
8,867 | 0.7% | Created | Attack recon | Undetected |
| 5 | 6468ac9f…34bf |
5,918 | 0.5% | Download | DDoS | Undetected* |
| 6 | d9c5bd8d…acf4 |
5,104 | 0.4% | Download | Attack dropper | Malicious |
| 7 | d46555af…a98e |
1,219 | 0.1% | Upload | Attack (crypto) | Malicious |
| 8 | 16174ef4…a080 |
1,130 | <0.1% | Download | Attack dropper | Malicious |
| 9 | 80c15d81…762d |
972 | <0.1% | Created | Attack dropper | Undetected |
| 10 | 32f2fa94…c864 |
743 | <0.1% | Download | Attack download | Undetected* |
Undetected identifies those samples that were not reported by either the official VirusTotal vendors or members of the wider VirusTotal community, which includes individual researchers and honeypot operators. Undetected* identifies samples that were not reported by the VirusTotal vendors but were identified as suspect or malicious by the VirusTotal community.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 1. Full SHA-256 hashes, for blocklists and lookups
With the help of a number of cyber analysis tools and services (VirusTotal, URLhaus and Joe Sandbox), and Artificial Intelligence (AI) large language models (LLMs) such as Anthropic Claude Sonnet 5.5 and SpaceXSI Grok 4.7, a few meaningful observations and insights can be deduced from the data:
-
Created files are few and heavily reused. Five redirection-type files (ranks 1 to 4 and 9) appear in 442,192 attacks. Because each use involves at least one creation event, they account for at least 98.8% of the 447,720 creations, while the other 604 created files share at most 5,528 events.
-
The rank 1 and 2 samples point to cryptojacking preparation, and both lead by a wide margin. In particular, they are associated with a long-lived Linux botnet built around Shellbot, a Perl-based Internet Relay Chat (IRC) bot, operated by the hacking group Outlaw, believed to be Romanian. The campaign is a cryptojacking operation (cryptomining done on someone else’s machine) that has remained active since at least mid-2018. The group originally spread the bot by exploiting a common command injection vulnerability on IoT devices and Linux servers, and later campaigns shifted toward SSH brute-force attacks against servers with weak credentials. Once a machine is infected, it takes commands over an IRC channel, including port scans, DDoS attacks and file downloads. The main goal is cryptojacking: the toolkit combines the IRC bot for command and control, Monero mining for revenue, and the haiduc scan and brute-force tool, and newer analysis says it also deploys modified XMRig miners and spreads in a worm-like way. XMRig is a popular open-source cryptocurrency mining software, primarily used to mine Monero (XMR) and other coins that use the RandomX algorithm. The rank 1 sample appears in 26.3% of active attacks and 19.2% of successful attacks, making it about seven times as common as the next sample. Together, the rank 1 and rank 2 samples appear in 30.0% of all active attacks, about ten times as often as the next sample.
-
The rank 1 sample is the hash of a malicious SSH
authorized_keysfile used as a persistence backdoor, commonly labeledtrojan.shell/malkey. Attackers brute-force or exploit weak or default SSH credentials on internet-facing Linux servers, IoT devices, containers and poorly secured hosts. Once they have a shell, they append (or replace) this specific public key in the victim’sauthorized_keysfile. The key carries the comment stringmdrfckr, and with the matching private key the attackers gain permanent passwordless SSH access, even if passwords are changed later. From that foothold they deploy coinminers (often XMRig-based) and IRC and Shellbot components, remove competing miners, and sometimes add further backdoors or scanners to spread. Related names and tooling include Outlaw, Shellbot and various XMRig droppers that plant this exact key. -
The rank 2 sample is the hash of a single newline character. When the honeypot receives a request for a file, such as those Shellbot makes, it often answers with an HTTP 200 OK status whose body is exactly one newline.
-
-
The rank 3 and 4 samples are indicators of cryptojacking reconnaissance. Cybercriminals often run automated reconnaissance to decide whether a host is worth further infection and which architecture-specific payload to fetch. They do this with a multi-stage fingerprinting script that tries several ways to get
unameoutput; determines the target architecture (x86_64, aarch64, armv7l and so on) with similar fallbacks; collects uptime, CPU count, CPU model, GPU information and recent logins; runs a shell-behavior test that creates a temporary script that echoes a string, makes it executable, runs it and deletes it; and prints everything with labeled prefixes. The rank 3 and rank 4 samples are the hashes of that temporary script or of its output, and they show the same labeled hardware-fingerprinting style documented in SSH bots that triage systems for mining suitability. Together they appear in about 3.5% of all active attacks. -
The rank 5 sample is the hash of a “The service is unavailable” response. It was seen in a DDoS attack against the website of the National Library of Israel.
-
The rank 6 sample is the hash of a Mirai dropper. It is associated with the file
ssh.sh, a well-known Mirai-related shell script that downloads architecture-specific payloads after a successful SSH login. Mirai is an IoT botnet that first appeared in 2016. It infects internet-connected devices such as cameras, DVRs and home routers, then uses them to launch large DDoS attacks. Infected devices scan the internet for others with open telnet (and later SSH) ports and try a built-in list of common default credentials, such as admin/admin, so devices whose owners never changed the factory password are easy prey. Bots connect to a command-and-control server and wait for attack orders. Mirai lives in memory and has no persistence, so a reboot clears the infection, but an exposed device with default credentials is usually reinfected within minutes, since its IP address was already captured when it was first infected. -
The rank 7 sample is the hash of a cryptojacking script. It is associated with
clean.sh, a short cleanup or uninstall-style script used by cryptojacking botnets. It targets the C3Pool miner (a legitimate-looking XMRig setup that is frequently abused), typically to remove competing miners or to clean up after an installation. -
The rank 8 sample is the hash of another Mirai dropper. In this case the file name is
gay.sh, which fits the pattern of Mirai-family botnets that deliberately use provocative or meme-style filenames. -
The rank 9 sample is the hash of a non-Mirai dropper. Specifically, it is a classic Gafgyt/Bashlite-style IoT botnet dropper, tried over SSH or Telnet against routers and embedded devices. Gafgyt is a long-running botnet, first seen in 2014, that, like Mirai, targets IoT devices in order to build a botnet for DDoS attacks. The observed tiny-dropper, multi-fallback-download and BusyBox pattern fits that lineage.
-
The rank 10 sample is the hash of the response to a file download request. Although this is the hash that uniquely identifies the default Nginx “403 Forbidden” HTML error page, automated analysis tools, such as the Hatching Triage Sandbox, list this hash when malware scripts attempt to retrieve secondary payloads or reach a Command & Control (C2) server.
-
Transfers are a long tail. The five most-used transferred files appear in only 14,114 attacks, set against 5.13 million transfer events and 1.84 million unique transferred files. Payloads are rarely repeated.
-
Verdict coverage is weak. By attack count, 77.7% of top-ten activity from 4 samples involves files rated malicious by VirusTotal, 12.0% activity from 3 samples as suspect (i.e., reported by the community, but not by the primary security vendors), and 10.3% activity from 3 samples are unreported.
-
Unreported (or even a harmless rate) does not mean benign. The rank 3, 4, and 9 samples appear as not reported in VirusTotal, yet they were created in 47,028 attack sessions. The extended VirusTotal community, but not the main vendors, reported the rank 2, 5, and 10 samples, which combined appeared in 54,752 attacks. These samples represent files that are not malicious per se (e.g., new line, HTTP status codes or messages), but are artifacts and indicators of attack activity. As such, they should be reviewed, not dismissed.
Because transferred files are almost never repeated, attackers likely vary their payloads per delivery. Hash-based blocking will therefore catch only a sliver of transfers, and behavioral and network indicators are more durable.
Report: 5.8 Million Knocks: A Three-Year Honeypot Record of SSH and Telnet Attacks
Leave a Reply