How it works¶
The idea¶
A honeypot is a fake system that exists only to be attacked. Real users never need it, so any connection to it is suspicious by definition. There is no normal business traffic to a decoy, so there are no false positives from it.
ReD-Pot places such decoys inside your network and makes them look like your servers. A decoy behaves like a real server up to the point of a successful login, and then serves believable but fake data. Nothing an attacker sends is executed. Every decoy carries bait (credentials, keys, documents), and each bait is a honeytoken with a unique id: when it is used anywhere, ReD-Pot knows where it was stolen from.
sequenceDiagram
participant WS as ACC-PC-07 (phished)
participant FS as RS-FS-01 (file server decoy)
participant WEB as RS-WEB-01 (web decoy)
participant M as ReD-Pot manager
WS->>FS: 1. scan, list shares
FS-->>M: L3 intent
WS->>FS: 2. read \\RS-FS-01\IT\passwords.txt
FS-->>M: L4 compromise
WS->>WEB: 3. SSH with the stolen deploy password
WEB-->>M: L5 honeytoken: "password planted on RS-FS-01 used on RS-WEB-01 by ACC-PC-07 (anna.k)"
M->>M: one incident, attack path FS-01 → WEB-01, alerts.json
Why internal deception works¶
| Stage | What the attacker does | Example |
|---|---|---|
| 1. Initial access | gets in | phishing e-mail, stolen VPN password, exposed service |
| 2. Foothold | establishes a stable presence on one machine | malware or a remote-access tool on a workstation |
| 3. Discovery | looks around the network | port scans, share enumeration, LDAP queries, DNS zone transfer |
| 4. Credential access | collects passwords | passwords.txt, .env files, Group Policy passwords, Kerberoasting |
| 5. Lateral movement | uses those credentials on other servers | SSH, RDP or SMB into the next machine |
| 6. Impact | steals data or deploys ransomware | the expensive part |
Perimeter controls and EDR focus on stage 1. Attackers who get past them often use legitimate tools and real credentials, so they look like normal admins. ReD-Pot targets stages 2 to 5, the window between "they are in" and "the damage is done": the attacker has to explore to move on, and the decoys sit exactly where they look.
Scoring¶
ReD-Pot scores how far a source got, not how many packets it sent. Events from one source are grouped into one incident that only goes up the ladder:
| Level | Name | Meaning |
|---|---|---|
| L1 | touch | TCP handshake only |
| L2 | interaction | protocol data exchanged |
| L3 | intent | trying to get in or map things |
| L4 | compromise | it worked: login, file read, command, tampering |
| L5 | honeytoken | a stolen bait was used |
Details and thresholds: Incidents and levels.
Who did it¶
ReD-Pot identifies the source as precisely as the protocols allow:
- Workstation name from SMB, RDP (CredSSP/NLA) and NTLM messages.
- Logged-on user: only an account a decoy really accepted.
- Users tried: every name typed, sprayed or sent.
- Tools: HTTP user agents, SSH client versions, the RDP cookie.
Hard to fingerprint¶
- Your domain, host names, users, files and certificates; passwords and file contents differ per installation.
- Protocol details follow the real products: SMB dialects and OS version of Windows Server 2019, RDP with NLA and a Remote-Desktop-style certificate, Windows DNS behaviour on the DC, Apache error pages on the web decoy, TTL 128 for Windows decoys.
- On a decoy's interface only that decoy's ports answer, and the decoy uses the interface's real MAC address.
Known limits of the Community Edition decoys
- The Windows decoys negotiate SMB 2.0.2 and 2.1 only. A client that lists dialects (for example
nmap --script smb-protocols) sees that SMB 3.x is missing. - Decoys use the Linux host's TCP/IP stack, so active OS fingerprinting (
nmap -O) reports Linux, also for a Windows persona. - Kerberos-authenticated (GSSAPI) LDAP binds are refused on the domain controller; NTLM binds work.