RedAmon: A new frontier in AI-powered Penetration Testing automation
- Thảo Nguyên
- Jun 30
- 4 min read
In the cybersecurity field, penetration testing - the method of simulating cyberattacks to identify vulnerabilities - has always been a time-consuming and labor-intensive process. The emergence of RedAmon, a new open-source cybersecurity platform, is bringing a fresh perspective to this workflow. The platform seamlessly bridges every stage of the process, from initial reconnaissance and exploitation to AI-driven vulnerability classification and automated remediation. Notably, this closed-loop workflow culminates in a ready-to-merge pull request containing the patch on GitHub.
RedAmon's flexible modular architecture
RedAmon is built on a modular architecture and runs entirely on Docker (a technology that encapsulates applications within isolated environments). This setup eliminates the need for users to directly install a barrage of complex security tools onto their host systems, minimizing the risk of software conflicts.
The system operates comprehensively based on six core pillars:
Parallelized reconnaissance pipeline: Parallel processing of information gathering and system scanning.
AI agent orchestrator: The orchestrating hub for artificial intelligence agents.
Attack surface graph: A visual mapping of the system's entire attack surface.
EvoGraph: The component responsible for data persistence and querying across sessions.
Cypherfix remediation engine: The central engine for automated incident handling and remediation.
Project settings engine: A project configuration manager that allows fine-tuning with over 500 different parameters.
Reconnaissance and attack surface analysis pipeline
The first stage of the attack chain is reconnaissance. RedAmon can concurrently trigger over 40 industry-standard security tools (such as Subfinder, Amass, Naabu, Masscan, Nuclei, Katana, FFuf, and Arjun). All of these tools run centrally inside a containerized Kali Linux environment.
Instead of forcing users to process individual output files, all generated data is synced directly into a unified Neo4j-based knowledge graph. Structurally sound with 17 node types and over 20 relationship types, this graph enables AI agents to gain a comprehensive understanding of the target's attack surface and execute complex queries in minutes rather than hours.
Expanding scanning capabilities to AI/LLM systems
Moving beyond traditional systems, RedAmon features a specialized module called AI Gauntlet, designed to assess the security of AI platforms and Large Language Models (LLMs).
This module orchestrates 4 dedicated red-teaming tools: garak, PyRIT, Giskard, and promptfoo. They simulate attacks against discovered endpoints to identify threats like prompt injection, jailbreaks, or data exfiltration. Any uncovered vulnerabilities are categorized and mapped against the OWASP-LLM and MITRE-ATLAS frameworks.
Autonomous AI agent-driven exploit management
At the core of RedAmon's orchestration is an autonomous agent built on LangGraph, operating under the ReAct (Reasoning + Acting) framework. This agent autonomously drives the attack lifecycle through three sequential phases: Reconnaissance, Exploitation, and Post-Exploitation. Utilizing open-source MCP (Model Context Protocol) servers running within a securely sandboxed Kali Linux environment, the AI agent commands over 14 specialized security tools.
The toolkit includes Metasploit for exploit execution, Hydra for brute-force credential attacks, Playwright for browser automation, alongside a full Kali command-line interface packed with over 70 pre-installed utilities.
Fireteam coordination mode
When Fireteam mode is activated, the primary AI agent can spawn multiple sub-agents to handle specialized tasks simultaneously. For instance, at any given moment, one sub-agent might utilize Hydra to audit password policies, another could chart a privilege escalation path from a specific CVE, while a third conducts an XSS vulnerability scan on the frontend user interface.
Automated source code patching and risk assessment
RedAmon sets itself apart from conventional scanners through its automated remediation capability, driven by the CypherFix system and a two-agent workflow:
Triage agent: Utilizes 9 predefined Cypher queries to scan the Neo4j knowledge graph. It connects and analyzes hundreds of discovered vulnerabilities, deduplicates results, and ranks them based on real-world exploitability.
CodeFix agent: Once triage results are ready, this agent clones the target's source code repository. Leveraging 11 source-code analysis tools, CodeFix locates the exact flaw, applies the fix within a ReAct loop, and automatically submits a pull request on GitHub for human review and approval.
Safety guardrails and the human-in-the-loop element
Despite its high level of automation, RedAmon is engineered to keep humans in ultimate control through its Tool Confirmation feature. The system establishes gating checkpoints, forcing the AI agent to pause and request operator permission before executing intrusive actions (such as Nmap scans, running Metasploit, or brute-forcing passwords with Hydra). Users can click "Allow" or "Deny" directly within a real-time timeline chat interface.
To guarantee legal compliance and safety, users can upload a Rules of Engagement (RoE) document, allowing the framework to automatically enforce project boundaries. Additionally, a platform-level Target Guardrail feature permanently blocks scanning against government, military, and educational domains.
The development team and AI model support
The RedAmon platform is developed and maintained by two experts:
Samuele Giampieri: An AWS-certified AI Platform Architect with over 15 years of experience building enterprise-grade AI agent systems.
Ritesh Gohil: A Cybersecurity Engineer at Workday with over 7 years of penetration testing experience and the discoverer of 11 published CVE vulnerabilities.
The framework offers exceptional flexibility, supporting over 400 different large language models to accommodate dynamic project requirements. It is fully compatible with major LLM providers such as OpenAI (GPT-5), Anthropic (Claude Opus 4.6), and AWS Bedrock, as well as locally hosted models via Ollama.
RedAmon stands out as a groundbreaking automated penetration testing solution. It leverages the power of AI to rapidly discover security weaknesses while bridging the gap between detection and remediation through automated source code patching. The project's source code is currently publicly available for the community on GitHub at: https://github.com/samugit83/redamon.
Comments