Beelzebub

Beelzebub is an open-source deception runtime framework that deploys adaptive decoy services to collect threat intelligence. It uses large language models to engage attackers in realistic interactions across various protocols, moving beyond passive honeypots to capture actionable tactics, techniques, and procedures. By simulating authentic environments, the framework encourages attackers to reveal their methods while keeping them isolated from production systems.

The software is written in Go and can be deployed via Docker, Kubernetes using Helm, or as a standalone binary. It provides a structured command line interface for running services, validating configurations, and managing plugins. Users can define decoy services using YAML files, allowing for the rapid deployment of new targets without writing custom code for every interaction.

Key features

• Adaptive deception engine using OpenAI or Ollama for real-time responses
• Low-code service definition via YAML configurations and regex matching
• Multi-protocol support for SSH, HTTP, TCP, TELNET, and MCP
• Extensible plugin system for custom command and HTTP response generators
• Observability integration with Prometheus metrics
• Event streaming via RabbitMQ for SIEM integration
• Kubernetes deployment support through Helm charts
• MCP decoy tools to detect prompt injection attacks against AI agents

The framework is designed for security researchers and organizations implementing preemptive cybersecurity measures. It utilizes a two-tier configuration system, separating global core settings from individual service definitions. This architecture allows for granular control over memory limits and logging per service. The inclusion of the Model Context Protocol (MCP) allows the framework to act as a honeypot for AI agents, alerting administrators when a prompt injection attack convinces an agent to invoke a restricted decoy tool. This provides a method for measuring attack surface metrics and collecting the exact malicious prompts used in guardrail bypass attempts.

Beelzebub serves as a research-oriented tool for detecting guardrail bypasses and collecting high-fidelity attack data through the deployment of deceptive infrastructure.