Αυτοματοποίηση ελέγχου διείσδυσης μέσω αρχιτεκτονικής πολλαπλών AI agents

Μπαλτζής, Δημήτριος

Penetration testing automation through multi-agent AI architecture

Master Thesis

Author

Μπαλτζής, Δημήτριος

Date

2026-07

Abstract

This thesis proposes, designs, and implements an automated penetration testing system based on a multi-agent AI architecture, which is subsequently evaluated in a controlled environment. The system leverages the Model Context Protocol (MCP) and the DeepSeek language model to coordinate two specialized agents: an Analyst Agent, responsible for passive reconnaissance and generating a structured attack plan, and a Pentester Agent, which handles exploit execution and vulnerability confirmation. The reconnaissance phase is implemented through a pipeline consisting of Wappalyzer, FastCVE, and ExploitDB, while execution relies primarily on Nuclei templates with a fallback path via ExploitDB. The system was evaluated in two controlled Docker environments, successfully confirming a path traversal vulnerability (CVE-2021-41773) and a remote code execution vulnerability via reverse shell (CVE-2017-5638). Results demonstrate that the system can reliably automate the full penetration testing pipeline with minimal human intervention, validating the effectiveness of the multi-agent role separation approach.

Postgraduate Studies Programme

Κυβερνοασφάλεια και Τεχνολογίες Τεχνητής Νοημοσύνης / MSc Cybersecurity & AI Technologies

Department

Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Πληροφορικής

Number of pages

Language

Greek

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/19726

Collections

Τμήμα Πληροφορικής

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα