Threat Research

We run honeypot sensors, analyse what they capture in an isolated lab, and share vetted indicators of compromise with the security community.

How Our Research Works

Real attacks, captured and analysed, with the results shared openly

1

Capture

Two sensors run continuously: a WordPress bait with deliberately outdated plugins, behind a web application firewall that retains full request bodies, and a Cowrie SSH/Telnet honeypot that records the commands attackers run and keeps every file they download or upload.

2

Analyse

Samples are examined offline in an isolated lab on a separate machine, reset to a clean snapshot after every run. We record what a sample tries to do: the names it resolves and the addresses it reaches for.

3

Share

Vetted indicators go to community threat-intelligence platforms. Write-ups, indicator lists and detection rules (YARA, Suricata, Sigma) are published openly under CC BY 4.0.

Published Findings

Each finding has a write-up, an indicator list and detection rules in our public repository.

  1. Published

    Behind the 404s: the akuma loader, a new bot and a second C2 confirmed live

    The loader host that answered our Telnet honeypot only with error pages finally served its script on 10 October 2026. Run in our isolated lab, it fetched an x86-64 bot nobody had uploaded, which connected to a C2 port nobody had listed; the server accepted it in a short live run. A YARA rule we wrote for the loader also matched a script from another host a week earlier: the same template and the same Azure delivery machines, with a different bot behind it. Bot, URL and C2 are on MalwareBazaar, URLhaus and ThreatFox from us, with YARA, Suricata and Sigma rules.

  2. Published

    A loader host that served us only errors, and its bot names it as the C2

    Since 9 October 2026 an Azure machine has been trying to fetch a loader script from one host onto our Telnet honeypot, and the host only ever answered with error pages. An older MIPS bot from the same host was on URLhaus. We ran it in our isolated lab, fully offline, and its only destination was that host on another port: a C2 no one had listed, now reported to ThreatFox and to the hosting provider.

  3. Published

    "Tadashi", an ARM IoT bot whose download and C2 sit on one host

    Two machines installed the same ARM bot on our Telnet honeypot on 9 October 2026, with cron and rc.local entries that bring it back after a reboot. It matches no known family. We ran it in our isolated lab, found its C2 on the same host that serves the download, reported it to ThreatFox, and confirmed the server live the next day.

  4. Published

    A Mirai kit that hides its C2 servers in XOR-encoded DNS records

    A Mirai kit delivered to our Telnet honeypot keeps its server addresses in the DNS A records of two .gd domains, XOR-encoded with a fixed key, so a sandbox that sinkholes DNS never sees a server. We recovered the key offline by feeding the bot documentation addresses, confirmed it on two CPU builds and against the live records, and then saw both decoded servers accept the bot in a short live run. Neither server was on a public feed; both are now on ThreatFox and reported to their provider.

  5. Published

    An Events Calendar attack tool that 6.17.3 stops, and why 6.17.3 still isn't enough

    An automated tool posted a comment carrying a serialized PHP object chain, the approach of CVE-2026-78265, to the Events Calendar 6.17.3 site on our WordPress honeypot. That version rejects it, so nothing ran. We published the tool's fingerprints, including a user agent with no public hits when we searched for it. A site on 6.17.3 is still exposed to two later bugs on the same comment path, so the version to be on is 6.17.4.1 or later.

  6. Published

  7. Published

  8. Published

  9. Published

  10. Published

  11. Published

    A broken LZRD build that sends its own encoded strings

    A Telnet client hit our honeypot with logins and commands that looked like noise. Decoded, they are the credential list and shell probe of an LZRD build, a Mirai variant. The bot sends its strings without decoding them first, so most of its logins could never work on a real device.

Where We Contribute

We share indicators on these community platforms as ksi_digital.

abuse.ch Community

ksi_digital

Our contributor profile across the abuse.ch platforms: ThreatFox, URLhaus and MalwareBazaar.

View public profile

ThreatFox

@ksi_digital

Command-and-control (C2) servers and domains.

View public profile

URLhaus

@ksi_digital

URLs serving malware payloads.

View public profile

MalwareBazaar

@ksi_digital

Malware samples captured by our sensors.

No public profile page

YARAify (YARAhub)

ksi_digital

YARA rules for malware families we documented. They run against every new MalwareBazaar and YARAify upload.

View public profile

AbuseIPDB

ksi_digital

Addresses seen brute-forcing and logging in to our honeypot.

View public profile

SANS ISC DShield

ksi_digital

Sensor logs contributed to the SANS Internet Storm Center's DShield network.

No public profile page

Spamhaus Threat Intel Community

KSI Digital

IP addresses of command-and-control (C2) servers.

No public profile page

CrowdSec

Submitted

Cowrie JSON parser

Our Security Engine shares attack signals with the CrowdSec network. We contributed a Cowrie JSON log parser and tests to a community Hub collection, which is awaiting review.

View submission

GitHub

ksi-digital

Write-ups, indicators and detection rules.

View on GitHub

Methodology and Ethics

How we handle what we capture

Capture over contact

Our sensors record what attackers send to them. We never connect to attacker infrastructure outside the lab, never reuse harvested credentials or keys, and never scan or probe third-party hosts.

Offline, isolated analysis

Samples are analysed only in an isolated lab with no route to any production network. Its egress is fail-closed through an encrypted tunnel: if the tunnel drops, nothing leaves.

C2 contact only when necessary

A command-and-control (C2) server is contacted only when necessary to confirm it is live, for a short, time-boxed window, with only that single destination reachable and all other traffic blocked.

No binaries published

We never publish malware binaries. Samples are referenced by SHA-256 hash, and scripts are defanged before publication.

Vetted, deduplicated indicators

Indicators are checked against existing feeds before submission, so we report what is new rather than duplicating others. Confidence reflects how each indicator was observed.

Corrections welcome

'Not found publicly' is not the same as 'novel'. We say when we have simply not located prior reporting, and we correct our work when we get something wrong.

Contact, Disclosure and Citation

Research contact

For questions about our findings, corrections, or related observations you would like to share, email us. You can also open an issue on the repository.

christophe@ksi-digital.comOpen an issue

Security contact

Our security contact details are published in a standard security.txt file (RFC 9116).

View security.txt

Citing our work

The repository includes a CITATION.cff file with citation details. Content is licensed CC BY 4.0, so you may reuse it with attribution.

View CITATION.cff