Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Inducing Personality in LLM-Based Honeypot Agents: Measuring the Effect on Human-Like Agenda Generation

Lewis Newsham, Ryan Hyland, Daniel Prince

TL;DR

This work introduces SANDMAN, a modular architecture that uses LLM-driven Language Agents to create deceptive, human-like entities for cyber defense. Central to the approach is inducing personality traits via the Five-Factor Model (OCEAN) using the MPI framework, and studying how these traits shape task planning and scheduling. The authors demonstrate statistically significant effects of induced personality on schedule duration and task frequency, validating a prompt-based control mechanism for persona-driven behaviour. The findings suggest that personality-informed planning can enhance the believability and effectiveness of cyber deception agents, with implications for defense strategies, scalability, and ethical deployment in isolated environments.

Abstract

This paper presents SANDMAN, an architecture for cyber deception that leverages Language Agents to emulate convincing human simulacra. Our 'Deceptive Agents' serve as advanced cyber decoys, designed for high-fidelity engagement with attackers by extending the observation period of attack behaviours. Through experimentation, measurement, and analysis, we demonstrate how a prompt schema based on the five-factor model of personality systematically induces distinct 'personalities' in Large Language Models. Our results highlight the feasibility of persona-driven Language Agents for generating diverse, realistic behaviours, ultimately improving cyber deception strategies.

Inducing Personality in LLM-Based Honeypot Agents: Measuring the Effect on Human-Like Agenda Generation

TL;DR

This work introduces SANDMAN, a modular architecture that uses LLM-driven Language Agents to create deceptive, human-like entities for cyber defense. Central to the approach is inducing personality traits via the Five-Factor Model (OCEAN) using the MPI framework, and studying how these traits shape task planning and scheduling. The authors demonstrate statistically significant effects of induced personality on schedule duration and task frequency, validating a prompt-based control mechanism for persona-driven behaviour. The findings suggest that personality-informed planning can enhance the believability and effectiveness of cyber deception agents, with implications for defense strategies, scalability, and ethical deployment in isolated environments.

Abstract

This paper presents SANDMAN, an architecture for cyber deception that leverages Language Agents to emulate convincing human simulacra. Our 'Deceptive Agents' serve as advanced cyber decoys, designed for high-fidelity engagement with attackers by extending the observation period of attack behaviours. Through experimentation, measurement, and analysis, we demonstrate how a prompt schema based on the five-factor model of personality systematically induces distinct 'personalities' in Large Language Models. Our results highlight the feasibility of persona-driven Language Agents for generating diverse, realistic behaviours, ultimately improving cyber deception strategies.

Paper Structure

This paper contains 27 sections, 1 figure, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs in Defensive Applications:
  4. Language Agents:
  5. Agent Architectures:
  6. Gray Agent Simulation:
  7. Deceptive Agent Architecture
  8. Persona-based Task Planning in LLMs
  9. Inducing Personality Types in LLMs
  10. Experimental Method
  11. Persona-based Task Selection
  12. Neutral Task Behaviour
  13. Induced Personality Experiments
  14. Discussion
  15. Future Work
  16. ...and 12 more sections

Figures (1)

  • Figure 1: Architecture for SANDMAN agents, inspired by CoALA framework sumers2023cognitive.