LEARNING CYBERATTACK PATTERNS WITH ACTIVE HONEYPOTS

Abstract

Honeypots can detect new attacks and vulnerabilities like zero-day exploits, based on an attacker’s behavior. Existing honeypots, however, are typically passive in nature and poor at detecting new and complex attacks like those carried out by state-sponsored actors. Deception is a commonly used tactic in conventional military operations, but it is rarely used in cyberspace. In this thesis, we implemented “active honeypots,” which incorporate deception into honeypot responses. In five phases of testing, we incorporated deception techniques such as fake files, defensive camouflage, delays, and false excuses into a Web honeypot built with SNARE and TANNER software, and an SSH honeypot built with Cowrie software. Our experiments sought to investigate how cyberattackers respond to the deception techniques. Our results showed that most attackers performed only vulnerability scanning and fingerprinting of our honeypots. Some appeared to be performing horizontal scanning, accessing both honeypots in the same phase. We found that the attackers were primarily non-interactive and did not respond to customized deception. We also observed that attackers who established a non-interactive session might be unable to exit the session without external intervention. Thus, we can delay to penalize these attackers. We also discovered that some attackers used unusual means of transferring files to the SSH server, and we recommend exploring how deception can be used against such techniques.