Every month we count what hits our own production API server: an ordinary VPS in Germany with SSH on port 22 (key-only) and nginx on ports 80 and 443. These are the October 2026 totals. Only aggregates are published, never an IP address.
| October 2026 | |
|---|---|
| Failed SSH logins | 41,780 (about 1,348 a day) |
| SSH connections for accounts that do not exist | 61,079 |
| Unsolicited web requests | 60,577 (about 1,954 a day) |
| Distinct attacking addresses | 5,316 |
| ...already on a threat-intel list we load | 20% of them, sending 55% of the traffic |
What the web probes were after
| Target | Requests | Share |
|---|---|---|
| .env files | 13,731 | 23% |
| WordPress & PHP apps | 8,105 | 13% |
| Config, backup & build files | 6,399 | 11% |
| Command injection & RCE payloads | 3,909 | 6.5% |
| Non-HTTP protocol probes (TLS, SOCKS, RDP…) | 2,931 | 4.8% |
| Cloud & SSH credentials, instance metadata | 2,652 | 4.4% |
The usernames SSH bots tried
root 13,164 · admin 1,340 · ubuntu 814 · user 480 · debian 293 · test 205 · logtriage 157 · deploy 153 · ftpuser 146 · oracle 119 (and 5,191 other names).
Where it came from
By country of the attacking addresses: United States 32%, Netherlands 10%, United Kingdom 8.3%, China 6.4%, Germany 6.3%.
Busiest networks: Google LLC (AS396982, 433 addresses), DigitalOcean (AS14061, 432 addresses), Microsoft Corporation (AS8075, 409 addresses), Hydra Communications Ltd (AS25369, 367 addresses), Tencent Building (AS132203, 325 addresses).
What the clients said they were
- Browser-like UA (real or spoofed): 25,985 requests (43%)
- Fake Googlebot (spoofed UA): 8,449 requests (14%)
- Research & attack-surface scanners (Censys, LeakIX…): 7,173 requests (12%)
- curl: 4,914 requests (8.1%)
- No user agent: 4,898 requests (8.1%)
Takeaways
- The single biggest web target was .env files: 23% of all unsolicited requests. If your server answers those paths, that is where to look first.
-
rootalone took 32% of the failed SSH logins. On a key-only server with root login off, like this one, these guesses cannot succeed: they are noise, not risk, until someone turns password login back on. - Threat-intel lists named only 20% of the attacking addresses, but those sent 55% of the traffic: lists catch the busy scanners, not the long tail.
- 8,449 requests claimed to be Googlebot from addresses outside Google's published crawler ranges. A Googlebot user agent proves nothing on its own.
The full report, with daily charts, every category, the most requested paths and a JSON download (CC BY 4.0): October 2026 edition.
How this post was made: a script computes the numbers from the server's nginx access log and SSH journal, and this text is generated from those numbers, not written by AI. Totals only: no IP address is published.
Top comments (0)