markcardiff.tech:/honeypot-reviews/cowrie-ssh-honeypot-baseline-2026-08-09-to-2026-08-15.html
root@markcardiff:~/reports$ ./render cowrie-baseline --public-safe

Cowrie SSH Honeypot Baseline

Public-safe test layout for weekly SSH honeypot reporting. Coverage is 9–15 August 2026 UTC. This page intentionally omits raw payload code, attacker key material, detailed transcripts, and sensitive operational details.

Coverage 9–15 Aug 2026 UTC
Posture public-safe
Status baseline layout test
30,777Sessions
1,485Unique source IPs
19,232Failed login attempts
9,695Accepted honeypot logins

Observed activity chain

internetSSH scanscredentialautomationLinuxreconaccessstagingcleanup/evasion Public view: trend, behavior themes, and defensive takeaways. Private transcripts and raw payloads remain local-only.

Summary

  • Our SSH honeypot observed 30,777 connection sessions from 1,485 unique source IPs during the reporting week.
  • Automated login activity was substantial, with 19,232 failed attempts and 9,695 accepted honeypot logins.
  • Post-login automation generated 11,467 command events. Most activity was short-lived; median observed session duration was 1.24 seconds.
  • The dominant behavior was commodity Linux reconnaissance followed by attempts to modify SSH access, stage files, or clean up competing processes.

Key observations

  • SSH automation using Go-based clients dominated the observed client-version strings.
  • Repeated command patterns attempted to inspect system details such as kernel, CPU, disk, memory, and network configuration.
  • Multiple sessions attempted SSH access manipulation, including removal/recreation of .ssh paths and insertion of attacker-controlled authorized keys. Public report redacts key material.
  • Captured file activity included Linux multi-architecture payload naming patterns and a small number of Windows-looking filenames, suggesting broad opportunistic tooling rather than careful target selection.
  • The most active source IP was also flagged by VirusTotal with multiple malicious detections, but attribution is not assessed from this data alone.

Activity metrics

MetricValue
Sessions30,777
Unique source IPs1,485
Failed login attempts19,232
Accepted honeypot logins9,695
Command events11,467
File download events1,480
File upload events96

Common behavior themes

ThemeMatching command events
recon6,643
destructive or evasion2,912
persistence1,470
mikrotik probe26
downloaders2

Representative observed commands

The following examples are sanitized and shortened where needed:

CategoryExamples
System reconnaissanceuname, /proc/cpuinfo, lscpu, df, free, whoami
SSH access manipulationattempts to recreate .ssh and write attacker-controlled authorized keys
Cleanup/evasionattempts to remove temporary scripts, kill processes, or modify host access-control files
Router/embedded probing/ip cloud print style command attempts observed in a small number of sessions

Selected public-safe indicators

These indicators are provided for defensive awareness. They should not be treated as attribution by themselves.

IndicatorContext
45.153.34.165939 sessions
91.92.47.55773 sessions
45.156.87.34773 sessions
77.239.124.249773 sessions
91.92.42.61773 sessions
http://5.182.210.174/ok5 URL observation(s)

Defensive takeaways

  • Internet-exposed SSH services continue to receive high-volume automated credential attacks and immediate post-login automation.
  • Defenders should monitor for short-lived SSH sessions followed by system reconnaissance, downloader usage, .ssh/authorized_keys writes, and process-kill cleanup patterns.
  • HASSH/client-version clustering can help separate commodity scanners from more interactive or bespoke activity.
  • Honeypot data is useful for trend tracking and detection engineering, but individual source IPs may represent compromised infrastructure, scanners, VPN/proxy nodes, or short-lived cloud hosts.

Limitations

  • This report summarizes honeypot telemetry only; it does not imply compromise of production systems.
  • VirusTotal enrichment was intentionally limited to conserve quota.
  • Public output omits raw payload code, full attacker keys, detailed session transcripts, and sensitive operational details.