Inside Our Honeypot: 877,000 Intrusion Attempts Captured in 16 Days
Not a vendor report with unverifiable global figures — the real data from a machine we deliberately exposed to watch, live, what automated attackers actually do on the Internet.
November 14, 2026 · SunuCyberSecurity team
events captured by our honeypot (SSH + Telnet) in 16 days
distinct attacking IP addresses — an average of 270 per day
of logins marked 'successful' — by the honeypot's design, not a real flaw
times the same reconnaissance command was the very first one typed
Source: our Cowrie honeypot — data from October 30 to November 14, 2026
A honeypot is a machine that poses as an ordinary server — here, a fake SSH and Telnet server, isolated from our production infrastructure — for the sole purpose of observing who tries to break in, and how. Unlike a scan or an audit, nobody is solicited: the attackers come on their own, exactly as they would against any machine exposed on the Internet. Between October 30 and November 14, 2026, this machine captured 877,315 events, from 3,306 distinct IP addresses — an average of 270 different attackers every day, with no advertising and no inbound link to this machine. This is the Internet's permanent background noise, not an exceptional spike.
The number that looks alarming, and why it isn't
Of the 77,110 login attempts recorded, 76,794 are marked "successful" — a 99.6% rate. That number could suggest a catastrophic flaw. It isn't one: a Cowrie honeypot is deliberately configured to accept almost any credentials presented, precisely to let the attacker in and observe what they do next, rather than blocking them at the door like a real server would. This rate doesn't measure a security flaw — it measures the honeypot's own behavior.
What is a real signal, on the other hand: almost none of the attackers ever check whether their credentials were correct before continuing — they chain attempts in a loop, by script, without ever pausing on a refusal. That's the most direct proof we're looking at mass automation, not individuals specifically targeting one machine.
Where the attacks come from (and what that doesn't prove)
The United States (651 distinct IPs), China (410), then Brazil and Russia (155 each) top the geographic ranking of attacking addresses.
Country of origin of attacking IP addresses (top 8)
Source: geolocation of the 3,306 distinct IPs captured by our honeypot
An important methodological point: an IP address's geolocation shows where the traffic originates, not the nationality or real location of whoever launched the attack. The large majority of this traffic comes from cloud hosting providers, VPNs, or already-compromised devices (routers, cameras, poorly secured servers) recruited into botnets — a given IP's country says almost nothing about who's behind the keyboard, or even whether anyone still is at this stage.
The most-tried credentials: the botnets' dictionary by rote
The most-tried username is, unsurprisingly, root (27,621 times), followed by admin (6,274) and ubuntu (5,088) — the latter gives away a specific targeting of cloud virtual machines, for which Ubuntu is the default image at most hosting providers. More unexpected: claude appears 459 times and postgres 415 times — proof that the dictionaries now used include names tied to AI tools and common databases, not just the classic "admin" and "root".
Most-tried username/password combinations
Source: login attempts marked 'successful' by the honeypot
A revealing technical detail: the pair 345gs5662d34 / 345gs5662d34 (identical username and password, a random string) appears 381 times. An identical, non-human username/password pair like this one is a known signature of certain automated scanners that test whether a service accepts any pair as long as it's symmetric — script behavior, not an attempt to guess a real password.
Once "inside": the IoT botnet signature
The command uname -s -v -n -r -m (which reveals the operating system and processor architecture) was the very first command typed 54,883 times — an almost systematic automatic-reconnaissance reflex before any other action. Next come probes typical of routers and connected devices: enable, system, shell (several thousand times each) — commands that only make sense if the attacker expects to be talking to a router or a camera, not a real Linux server.
The most telling pattern remains the write attempt, repeated 1,725 times each, into six different directories (/var, /home, /data, /dev, /etc, /var/run: create a file, make it executable, run it, then try elsewhere if the first directory is read-only. This is the signature behavior of Mirai-derived botnet families, which target poorly secured connected devices — not a bespoke attack, but a script that systematically tries until it finds a writable location.
The malicious payloads observed
5,080 file download attempts were captured. A single server served, under different names, binaries compiled for eight different processor architectures (x86_64, aarch64, mips, mipsel, powerpc, sparc, sh4, m68k) — proof that a single botnet indiscriminately targets servers, home routers and connected devices regardless of architecture, letting the infected machine pick the right binary itself. We aren't publishing the exact addresses observed: they change constantly, and republishing them would only index already-ephemeral infrastructure, with no benefit to the reader.
What this means in practice
Never expose Telnet to the Internet
No configuration makes it safe — the protocol itself transmits credentials in plain text. 188,000 of the 877,000 captured events targeted Telnet: the traffic exists and will never stop on its own.
Disable password-based SSH login
root/admin, admin/admin, 123456... these credentials aren't guessed at random, they're tested in a loop by scripts. Public-key authentication makes this entire attack class automatically ineffective, with no extra configuration.
Change every default password on a connected device
The reconnaissance signature observed (uname, busybox probing, write attempts into /var, /dev, /etc) matches the classic behavior of botnets targeting connected devices (routers, cameras, NAS) still running factory credentials.
Don't assume silence means safety
If your SSH/Telnet service is exposed, this background traffic already exists against it, whether a security tool flags it or not — it's permanent Internet noise, not a rare event specifically targeting your organization.
Limit the surface, not just harden it
A firewall restricting SSH access to a known list of IPs (VPN, office) eliminates nearly all of this noise — far more effective than trying to filter attempts after they arrive by the thousand.
Methodology
Data captured by a deployment of Cowrie, a well-known open-source honeypot in the security community, exposing SSH and Telnet on an isolated machine with no access to our production systems. IP geolocation via the GeoLite2 database. Data window: October 30 to November 14, 2026 (16 days). These figures reflect what a machine exposed with no advertising receives as traffic — not a representative sample of all attacks targeting Africa or the world, only of what targets a generic exposed surface on the Internet.
Also worth reading
Check your own exposure
A free scan, in a few minutes, no credit card needed.
Free scanNewsletter
Get our next analyses
One email whenever we publish a new article. No spam, one-click unsubscribe.