Description
As computer networks grow in size, maintaining an accurate overview of network infrastructure becomes
increasingly challenging, especially with the dynamic nature of modern networks. This work introduces
ANMAD, the Automated Network Mapper and Asset Discovery framework designed to assist network
professionals.
ANMAD comprises two variants: ANMADv1 and ANMADv2. ANMADv1 attempts to fully automate network
mapping using a Large Language Model (LLM) backbone. This variant explores the feasibility of fully
autonomous AI agents for network exploration that do not require human intervention. ANMADv2 adopts
a hybrid approach, selectively integrating LLMs to enhance existing static mapping tools, and extends
mapping capabilities with active monitoring and incident response support.
Both variants underwent comprehensive evaluation across diverse network configurations. Results show
that while ANMADv1 exhibits potential in small-scale scenarios, it faces challenges in scalability and efficiency
for larger networks. This underlines the current impracticability of fully autonomous AI agent
solutions. In contrast, ANMADv2 provides superior performance and scalability in real-world applications.
ANMADv2 demonstrates particular efficacy in its extended capabilities, leveraging LLMs for reasoning and
dynamic network event management.
This research highlights the practical application of AI-assisted tools in network management, paving
the way for future advancements in autonomous systems and network security.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching